{ "referencePoints": { "schemaVersion": 1, "forecastSha256": "662a4369b05520d70255c65d8a52da1eee57923ffc0d171d53e3e08c234a568a", "ledgerSha256": "f03c7712c2be990c757e05422740dacafa4bc15d23abf227c516e6bb57256188", "updatedAt": "2026-09-26T21:05:49.181Z", "sources": { "ref-b5286382df13": { "url": "https://www.nih.gov/news-events/nih-research-matters/brain-computer-device-helps-man-speak", "urls": [ "https://www.nih.gov/news-events/nih-research-matters/brain-computer-device-helps-man-speak" ], "title": "Brain-computer device helps man speak", "organization": "National Institutes of Health", "quality": "official-agency", "transport": "browser", "format": "html", "publishedAt": "2026-07-14", "dateEvidence": "July 14, 2026", "retrievedAt": "2026-09-05", "reuse": [ { "family": "implantable-neural-communication", "domains": [ "technology" ], "ids": [ "horizon-implantable-neural-symbiosis" ] } ], "identity": "3ab73329dc09922e3cc077924500a3417171b886c5ffe5f00d934c4330f4fe60", "reviewSha256": "1341ab73a8594e92fb657eb563b9e53c60314468a789dd4c6fefa2e8c6fd5cb5", "health": { "status": "unavailable", "lastCheckedAt": "2026-09-26T21:04:24.424Z", "lastVerifiedAt": "2026-09-07T22:06:37.306Z", "textSha256": "a88ac0264d9978568b7ef1375cd2ff8b35d8073518a5539dad16e33169c3304c", "reviewSha256": "1341ab73a8594e92fb657eb563b9e53c60314468a789dd4c6fefa2e8c6fd5cb5", "error": "Browser source refused: HTTP 403", "etag": null, "lastModified": null, "retryAt": null } }, "ref-47e27d029551": { "url": "https://news.utexas.edu/2023/05/01/brain-activity-decoder-can-reveal-stories-in-peoples-minds/", "urls": [ "https://news.utexas.edu/2023/05/01/brain-activity-decoder-can-reveal-stories-in-peoples-minds/" ], "title": "Brain Activity Decoder Can Reveal Stories in People’s Minds", "organization": "The University of Texas at Austin", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2023-05-01T15:01:19+00:00", "dateEvidence": "2023-05-01T15:01:19+00:00", "retrievedAt": "2026-09-05", "reuse": [ { "family": "noninvasive-neural-communication", "domains": [ "technology" ], "ids": [ "horizon-non-invasive-neural-symbiosis" ] } ], "identity": "187ee7bd7da41ac996c652bf4abbc17b1922fcc5636a6858dfff3d1133ea841b", "reviewSha256": "b00609c936d7b48d811ffea2c93d5e5375e671c22819f4b896e95e7fbb4b0ee8", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:24.980Z", "lastVerifiedAt": "2026-09-26T21:04:24.980Z", "textSha256": "efcedd4e2fc046dfeb9a2055157ab44dbf6ad1493ae9f6df015255c2a132aab7", "reviewSha256": "b00609c936d7b48d811ffea2c93d5e5375e671c22819f4b896e95e7fbb4b0ee8", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-98d256433085": { "url": "https://flywire.ai/", "urls": [ "https://flywire.ai/" ], "title": "FlyWire", "organization": "FlyWire Consortium", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05", "reuse": [ { "family": "brain-structure-reconstruction", "domains": [ "individual" ], "ids": [ "horizon-whole-brain-emulation-and-uploading" ] } ], "identity": "8c3dc85633ce0432d94840fe326904138c8733d04309425892c9e2cf9e9a7a21", "reviewSha256": "393f098e01dce55f1ad5ffc9d4ea7a1edf73e48b885c9f5f479d20acc5cabd02", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:25.595Z", "lastVerifiedAt": "2026-09-26T21:04:25.595Z", "textSha256": "ed42146f242308a0223688c3690cbec5995e5aad7e13a497b7e0e5b57577a15e", "reviewSha256": "393f098e01dce55f1ad5ffc9d4ea7a1edf73e48b885c9f5f479d20acc5cabd02", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-fdc33c23511b": { "url": "https://www.caltech.edu/about/news/space-solar-power-project-ends-first-in-space-mission-with-successes-and-lessons", "urls": [ "https://www.caltech.edu/about/news/space-solar-power-project-ends-first-in-space-mission-with-successes-and-lessons" ], "title": "Space Solar Power Project Ends First In-Space Mission with Successes and Lessons", "organization": "California Institute of Technology", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2024-01-16T09:04:01-08:00", "dateEvidence": "2024-01-16T09:04:01-0800", "retrievedAt": "2026-09-05", "reuse": [ { "family": "orbital-power-infrastructure", "domains": [ "technology" ], "ids": [ "horizon-orbital-compute-to-proto-dyson" ] } ], "identity": "a876fac31366aa67e9687e0b90bd5ed5f689e61e58ca85515ac60365fd15c377", "reviewSha256": "21a1faedd7f69b0f2820cb37f0723af89a55949599d68b32516464d2a46f12d8", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:26.839Z", "lastVerifiedAt": "2026-09-26T21:04:26.839Z", "textSha256": "b80c3672b755c17ab43e146b41b7bce069ef8529f369600034d75587edce5f49", "reviewSha256": "21a1faedd7f69b0f2820cb37f0723af89a55949599d68b32516464d2a46f12d8", "error": null, "etag": null, "lastModified": "Sat, 26 Sep 2026 21:04:27 GMT", "retryAt": null } }, "ref-37625c812bb6": { "url": "https://www.energyinst.org/statistical-review", "urls": [ "https://www.energyinst.org/statistical-review" ], "title": "Statistical Review of World Energy", "organization": "Energy Institute", "quality": "official-research", "transport": "browser", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05", "reuse": [ { "family": "civilizational-energy-baseline", "domains": [ "technology" ], "ids": [ "horizon-kardashev-energy-scaling" ] } ], "identity": "85eacfea5940a7288e2c281449c93f9bd8a779c79e04454ae5f73fcd9024081e", "reviewSha256": "0193578ef7ceeadc4cc451ca3be33b85ea80fd93e11c39133d8951d7ce306e2a", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:32.559Z", "lastVerifiedAt": "2026-09-26T21:04:32.559Z", "textSha256": "789422125a7e767f260f2ca35b79ee35bc4c73374c7ff0ffc66f875c1f98032f", "reviewSha256": "0193578ef7ceeadc4cc451ca3be33b85ea80fd93e11c39133d8951d7ce306e2a", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-b5102ffbbc87": { "url": "https://arxiv.org/abs/quant-ph/9908043", "urls": [ "https://arxiv.org/abs/quant-ph/9908043" ], "title": "Ultimate physical limits to computation", "organization": "arXiv", "quality": "author-manuscript", "transport": "https", "format": "html", "publishedAt": "1999-08-13", "dateEvidence": "13 Aug 1999", "retrievedAt": "2026-09-05", "reuse": [ { "family": "physical-computation-bounds", "domains": [ "technology" ], "ids": [ "horizon-transcension-hypothesis" ] } ], "identity": "a7239009cae56f4b6cf5e0a57f4458515ffdf91befe35f0b8f87c0d26a2b39ba", "reviewSha256": "d7479d205ceabda7bea0f6df8a3e09d41d234f805d2941ba018a6fcc5180442c", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:32.604Z", "lastVerifiedAt": "2026-09-05T00:45:40.985Z", "textSha256": "c29661d9db6fc603dff0969c4c1a700d679a92abe54b5f649a3af1a756f24fbe", "reviewSha256": "d7479d205ceabda7bea0f6df8a3e09d41d234f805d2941ba018a6fcc5180442c", "error": null, "etag": null, "lastModified": "Mon, 24 Aug 2026 18:37:40 GMT", "retryAt": null } }, "ref-a114147745ae": { "url": "https://writings.stephenwolfram.com/2021/11/the-concept-of-the-ruliad/", "urls": [ "https://writings.stephenwolfram.com/2021/11/the-concept-of-the-ruliad/" ], "title": "The Concept of the Ruliad", "organization": "Stephen Wolfram Writings", "quality": "theory-proposal", "transport": "https", "format": "html", "publishedAt": "2021-11-10", "dateEvidence": "2021/11/10", "retrievedAt": "2026-09-05", "reuse": [ { "family": "ruliad-formalism-and-testability", "domains": [ "technology" ], "ids": [ "horizon-ruliad-testable-physics" ] } ], "identity": "aac344dc41c320b20797085e887e6ff83096cd94c7a81cd4c2eb3bf72242954d", "reviewSha256": "1f4541b1f75d44f55bfe89ab3af507e4050ee4a7bccf4284ff81d4379b1ef923", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:34.261Z", "lastVerifiedAt": "2026-09-26T21:04:34.261Z", "textSha256": "d9e277acc0cf61f76099422cb3df4aed302c4a3af6538ade4e8347716a249d1b", "reviewSha256": "1f4541b1f75d44f55bfe89ab3af507e4050ee4a7bccf4284ff81d4379b1ef923", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-5ad436675449": { "url": "https://www.figure.ai/news/production-at-bmw", "urls": [ "https://www.figure.ai/news/production-at-bmw" ], "title": "F.02 Contributed to the Production of 30,000 Cars at BMW", "organization": "Figure AI", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2025-11-19", "dateEvidence": "November 19, 2025", "retrievedAt": "2026-09-05T00:33:33.717Z", "reuse": [ { "family": "humanoid-automotive-sheet-metal-loading", "domains": [ "technology" ], "ids": [ "2026-3" ] } ], "identity": "e35e3b49473a310ab8ece96d857f790967b67f65ae7270477706f4aab7fb20a8", "reviewSha256": "9511c25cfdf33bf00eb53691866c03b04561978b3018cb23e2783fdc7e4950de", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:34.777Z", "lastVerifiedAt": "2026-09-26T21:04:34.777Z", "textSha256": "69bbf58328ef555868dc1c1b88e48f412562738d90ce02a004701de5fab19ea0", "reviewSha256": "9511c25cfdf33bf00eb53691866c03b04561978b3018cb23e2783fdc7e4950de", "error": null, "etag": "W/\"16nvg11lc84wos\"", "lastModified": null, "retryAt": null } }, "ref-da3f53e4c6fc": { "url": "https://clinicaltrials.gov/api/v2/studies/NCT06429735", "urls": [ "https://clinicaltrials.gov/api/v2/studies/NCT06429735" ], "title": "PRIME: An Early Feasibility Study of a Precise Robotically Implanted Brain-Computer Interface for the Control of External Devices", "organization": "U.S. National Library of Medicine / ClinicalTrials.gov; sponsor-submitted Neuralink record", "quality": "official-agency", "transport": "https", "format": "json", "publishedAt": "2024-05-28", "dateEvidence": "\"studyFirstPostDateStruct\":{\"date\":\"2024-05-28\",\"type\":\"ACTUAL\"}", "retrievedAt": "2026-09-05T00:33:33.721Z", "reuse": [ { "family": "intracortical-bci-early-feasibility-safety", "domains": [ "technology" ], "ids": [ "2026-7" ] } ], "schema": "clinicaltrials-v2", "identity": "88e73b0003816b01952a25bff7bc55b1b972f8f5fef59ed11f90cd496ccaeb75", "reviewSha256": "ab2998467035a63e3df6dd0ce1a9e74332f7cc981c1406854c425192ff5242d1", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:35.082Z", "lastVerifiedAt": "2026-09-25T21:04:28.225Z", "textSha256": "adf7d6c4161c75e00cea4aadfdcd726594a26abe9fd1ee0cd03d07a58bc5ca58", "reviewSha256": "ab2998467035a63e3df6dd0ce1a9e74332f7cc981c1406854c425192ff5242d1", "error": null, "etag": "\"0/0.34.1/mugqcr34\"", "lastModified": null, "retryAt": null } }, "ref-3b910b5ced30": { "url": "https://www.starcloud.com/starcloud-1", "urls": [ "https://www.starcloud.com/starcloud-1" ], "title": "Starcloud-1", "organization": "Starcloud", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:33:33.520Z", "reuse": [ { "family": "orbital-gpu-named-workload-demonstrator", "domains": [ "technology" ], "ids": [ "2026-8" ] } ], "identity": "8bb27f92aa2342dcba44db6356da6f371024a7cbbeb2ac3379da10412aee2851", "reviewSha256": "e590af066d8c4bf5637e3d534b550e86a0497d209c8cc2e36c2ed4acefce492d", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:35.561Z", "lastVerifiedAt": "2026-09-25T21:04:28.768Z", "textSha256": "8d6fe979b6427c0ea3d1fa6f6209900f5819207aa65b44ef7cf14e1c30351114", "reviewSha256": "e590af066d8c4bf5637e3d534b550e86a0497d209c8cc2e36c2ed4acefce492d", "error": null, "etag": null, "lastModified": "Fri, 25 Sep 2026 06:34:24 GMT", "retryAt": null } }, "ref-7a179a48aad8": { "url": "https://www.iea.org/reports/energy-and-ai/executive-summary", "urls": [ "https://www.iea.org/reports/energy-and-ai/executive-summary", "https://www.iea.org/reports/energy-and-ai" ], "title": "Energy and AI — Executive summary", "organization": "International Energy Agency", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2025-04-10", "dateEvidence": "10 April 2025", "retrievedAt": "2026-09-05T00:33:33.491Z", "reuse": [ { "family": "datacenter-capital-and-electricity-buildout", "domains": [ "geopolitical", "economic" ], "ids": [ "2027-5", "2028-4", "2030-5", "2034-2" ] } ], "dateEvidenceUrl": "https://www.iea.org/reports/energy-and-ai", "identity": "da46720ce54949047555007d80d2660186cdff281874bfd7efb6567fa205ca62", "reviewSha256": "92f8ff28fb8ce13c86a1e420b30188366d64953558b542c14a0099e68a7a9a77", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:36.012Z", "lastVerifiedAt": "2026-09-26T21:04:36.012Z", "textSha256": "2ddfbe56107c7c1121c0ab099909f111c56e5a98d1d38442faee40680644dd42", "reviewSha256": "92f8ff28fb8ce13c86a1e420b30188366d64953558b542c14a0099e68a7a9a77", "error": null, "etag": null, "lastModified": "Sat, 26 Sep 2026 20:03:56 GMT", "retryAt": null } }, "ref-fea5cc03eeb8": { "url": "https://mrclmb.ac.uk/news-events/articles/complete-synaptic-resolution-connectome-of-an-insect-larval-brain/", "urls": [ "https://mrclmb.ac.uk/news-events/articles/complete-synaptic-resolution-connectome-of-an-insect-larval-brain/" ], "title": "Complete synaptic-resolution connectome of an insect larval brain", "organization": "MRC Laboratory of Molecular Biology", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2023-03-10", "dateEvidence": "Published 10/03/2023", "retrievedAt": "2026-09-05T00:50:51.848Z", "reuse": [ { "family": "larval-whole-brain-synaptic-reconstruction", "domains": [ "technology" ], "ids": [ "2028-7" ] } ], "identity": "84e7f3f659938c57181958d21337361559d0d308a0f07d0edf8725648e2abee6", "reviewSha256": "4d115d51199c2cbf084a235a6c5b763971a62b74a9bcfe7e80257683b1c51892", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:37.330Z", "lastVerifiedAt": "2026-09-26T21:04:37.330Z", "textSha256": "5c43de2d113ec3f85f1812ddd4cbada909acd132ceeed61350d2851d17a1bc41", "reviewSha256": "4d115d51199c2cbf084a235a6c5b763971a62b74a9bcfe7e80257683b1c51892", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-16ebffb99c27": { "url": "https://insilico.com/news/tnrecuxsc1-insilico-announces-nature-medicine-publi", "urls": [ "https://insilico.com/news/tnrecuxsc1-insilico-announces-nature-medicine-publi" ], "title": "Insilico Announces Nature Medicine Publication of Phase IIa Results of Rentosertib, the Novel TNIK Inhibitor for Idiopathic Pulmonary Fibrosis Discovered and Designed with a Pioneering AI Approach", "organization": "Insilico Medicine", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2025-06-03", "dateEvidence": "2025-06-03 10:05", "retrievedAt": "2026-09-05T00:50:50.262Z", "reuse": [ { "family": "ai-designed-small-molecule-clinical-development", "domains": [ "technology" ], "ids": [ "2030-4" ] } ], "identity": "2fb2ae5700571fb67befbdb4a8bfe2d6025505b32205ba49c7f08d1d12746244", "reviewSha256": "b25e5142daa06a6dce5424c79db7fbe5e46a9653b583f2c7a11c1122e5db4917", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:38.064Z", "lastVerifiedAt": "2026-09-26T21:04:38.064Z", "textSha256": "ec4ae049f643aa77e97431b2e06ad33158b42125f292dee7f8bba6e16df30d56", "reviewSha256": "b25e5142daa06a6dce5424c79db7fbe5e46a9653b583f2c7a11c1122e5db4917", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-bb9bfbd05aa4": { "url": "https://behavior.stanford.edu/", "urls": [ "https://behavior.stanford.edu/" ], "title": "BEHAVIOR", "organization": "Stanford University BEHAVIOR research team", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:33:33.484Z", "reuse": [ { "family": "human-relevant-physical-task-evaluation", "domains": [ "technology" ], "ids": [ "2032-1" ] } ], "identity": "6ff349bc6de5e14b34d011519d1a6cf509b59bbf9dd8c0f34c8eb49d6a859126", "reviewSha256": "72c61232f5d3f71a993ba099e1db3cc08c6b12cba33077f6c11506e617da05a9", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:38.122Z", "lastVerifiedAt": "2026-09-26T21:04:38.122Z", "textSha256": "99c66c88c2e5c1dbc5a909e41a3931709676d1ee82ba565f7fcd50466d006b2c", "reviewSha256": "72c61232f5d3f71a993ba099e1db3cc08c6b12cba33077f6c11506e617da05a9", "error": null, "etag": "W/\"6ab79c3f-2bba8\"", "lastModified": "Sat, 26 Sep 2026 10:19:43 GMT", "retryAt": null } }, "ref-e9b8dee2be94": { "url": "https://www.iea.org/reports/global-critical-minerals-outlook-2025/executive-summary", "urls": [ "https://www.iea.org/reports/global-critical-minerals-outlook-2025/executive-summary", "https://www.iea.org/reports/global-critical-minerals-outlook-2025" ], "title": "Global Critical Minerals Outlook 2025 — Executive summary", "organization": "International Energy Agency", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2025-05-21", "dateEvidence": "21 May 2025", "retrievedAt": "2026-09-05T00:33:33.510Z", "reuse": [ { "family": "critical-mineral-investment-and-supply-response", "domains": [ "economic" ], "ids": [ "2032-3", "2036-4" ] } ], "dateEvidenceUrl": "https://www.iea.org/reports/global-critical-minerals-outlook-2025", "identity": "555bf3cdd95d061871cbdb157f62c639a481c6d026f2f9f2ad6a2e71ab66caf7", "reviewSha256": "c5360fd0b6f7a90aed205bb0cdd6e4cc8b71154bf6124a07104d1c85af2f0b39", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:41.745Z", "lastVerifiedAt": "2026-09-26T21:04:41.745Z", "textSha256": "95b0fb58ce54704a61e615f381df52f592d24d7097fa08293dfc5ee93a66a546", "reviewSha256": "c5360fd0b6f7a90aed205bb0cdd6e4cc8b71154bf6124a07104d1c85af2f0b39", "error": null, "etag": null, "lastModified": "Sat, 26 Sep 2026 21:04:40 GMT", "retryAt": null } }, "ref-20c7e85747cb": { "url": "https://www.cdc.gov/wastewater/about/", "urls": [ "https://www.cdc.gov/wastewater/about/" ], "title": "About CDC's Wastewater Monitoring Program", "organization": "U.S. Centers for Disease Control and Prevention", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2026-04-10", "dateEvidence": "April 10, 2026", "retrievedAt": "2026-09-05T00:33:33.700Z", "reuse": [ { "family": "publicly-funded-wastewater-pathogen-surveillance", "domains": [ "governance" ], "ids": [ "2033-6" ] } ], "identity": "6970219e78b4bfb593136e92edb70c6739e1a083f99f112caefdfe554c9ab612", "reviewSha256": "2a7966725c8edd6ab76fd32eef64f8894b6d0870530b118cd3b3fae99121bd6f", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:42.527Z", "lastVerifiedAt": "2026-09-26T21:04:42.527Z", "textSha256": "0519f30aa02328007a8052e4307290331fbbe8a75713e94a5998f762e45ba1e1", "reviewSha256": "2a7966725c8edd6ab76fd32eef64f8894b6d0870530b118cd3b3fae99121bd6f", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-78446ff77822": { "url": "https://deepmind.google/blog/how-alphachip-transformed-computer-chip-design/", "urls": [ "https://deepmind.google/blog/how-alphachip-transformed-computer-chip-design/" ], "title": "How AlphaChip transformed computer chip design", "organization": "Google DeepMind", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2024-09-26", "dateEvidence": "September 26, 2024", "retrievedAt": "2026-09-05T00:33:34.715Z", "reuse": [ { "family": "ai-semiconductor-floorplanning-deployment", "domains": [ "technology" ], "ids": [ "2034-0" ] } ], "identity": "5e3ceb4349b417a12ce00f66844280d398c36a457789d087c9de08d9f0e8d9f9", "reviewSha256": "30c145d2caddd74f0bcb4869b594f1e89fdb35c909a3b60596901f9e05a08ba2", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:43.848Z", "lastVerifiedAt": "2026-09-26T21:04:43.848Z", "textSha256": "2527c6a9c1f787f324a1e13310e16e8af762a1e3188f62fbb99ad60f5d1a11bd", "reviewSha256": "30c145d2caddd74f0bcb4869b594f1e89fdb35c909a3b60596901f9e05a08ba2", "error": null, "etag": "\"GsuqZA\"", "lastModified": null, "retryAt": null } }, "ref-2cd5a560d120": { "url": "https://news.microsoft.com/source/features/sustainability/project-natick-underwater-datacenter/", "urls": [ "https://news.microsoft.com/source/features/sustainability/project-natick-underwater-datacenter/" ], "title": "Microsoft finds underwater datacenters are reliable, practical and use energy sustainably", "organization": "Microsoft", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2020-09-14", "dateEvidence": "2020-09-14T07:00:00+00:00", "retrievedAt": "2026-09-05T00:33:33.518Z", "reuse": [ { "family": "offshore-sealed-datacenter-feasibility", "domains": [ "geopolitical" ], "ids": [ "2034-4" ] } ], "identity": "73ac9a915b01b83e1e8dd18eeb8b42c75c0e8b1abcfbbd5ae02fc64883d29659", "reviewSha256": "5d694da7edcf7adc7a53ed63cc2498e4c3b8461f925a8b835a95ed603ae8b43e", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:44.105Z", "lastVerifiedAt": "2026-09-05T01:07:23.372Z", "textSha256": "3634658d50d64cc177aec3ec8c7b5fa5edac047eeb6475a0009008bc98c42b8d", "reviewSha256": "5d694da7edcf7adc7a53ed63cc2498e4c3b8461f925a8b835a95ed603ae8b43e", "error": null, "etag": null, "lastModified": "Mon, 24 Jul 2023 13:25:17 GMT", "retryAt": null } }, "ref-08731f05d093": { "url": "https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years", "urls": [ "https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years" ], "title": "World Robotics 2025 report – INDUSTRIAL ROBOTS – released by IFR", "organization": "International Federation of Robotics", "quality": "primary-reporting", "transport": "https", "format": "html", "publishedAt": "2025-09-25", "dateEvidence": "Frankfurt, Sep 25, 2025", "retrievedAt": "2026-09-05T00:47:27.963Z", "reuse": [ { "family": "industrial-robot-operational-stock", "domains": [ "economic" ], "ids": [ "2036-0" ] } ], "identity": "95ced6ae7eb90013fd01c9eb819b5b7adb74cd4b24dcdf2510b9cf6ede576150", "reviewSha256": "c63148fe7b1a28888acfdf635093eba6c8d249c9e97b66c3e50fccbdf25f7340", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:44.612Z", "lastVerifiedAt": "2026-09-26T21:04:44.612Z", "textSha256": "28e823cfdf90f6bcc2954fbfcd005513dbc24372d6c1d1c7b791b5803f7a9562", "reviewSha256": "c63148fe7b1a28888acfdf635093eba6c8d249c9e97b66c3e50fccbdf25f7340", "error": null, "etag": null, "lastModified": "Sat, 26 Sep 2026 21:04:45 GMT", "retryAt": null } }, "ref-d0f5ca5d8a30": { "url": "https://deepmind.google/blog/gemini-robotics-brings-ai-into-the-physical-world/", "urls": [ "https://deepmind.google/blog/gemini-robotics-brings-ai-into-the-physical-world/" ], "title": "Gemini Robotics brings AI into the physical world", "organization": "Google DeepMind", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-03-12", "dateEvidence": "March 12, 2025", "retrievedAt": "2026-09-05T00:33:34.768Z", "reuse": [ { "family": "embodied-robot-generality-and-dexterity", "domains": [ "technology", "economic" ], "ids": [ "2036-1", "2040-0" ] } ], "identity": "95fa786a7aee260da47873082765be8cbe52f449b99a349d71944dd899e92db8", "reviewSha256": "3f8e016d20805b92088e4b142f3365c4e2ba97b17a79a0cf1c39a735041b0efd", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:45.883Z", "lastVerifiedAt": "2026-09-26T21:04:45.883Z", "textSha256": "b2f5471dd13172fd71ce38f8a73a9eb12504393edaa9d2d0ace59ab0a0f10a27", "reviewSha256": "3f8e016d20805b92088e4b142f3365c4e2ba97b17a79a0cf1c39a735041b0efd", "error": null, "etag": "\"GsuqZA\"", "lastModified": null, "retryAt": null } }, "ref-273d762a7c91": { "url": "https://deepmind.google/blog/accelerating-fusion-science-through-learned-plasma-control/", "urls": [ "https://deepmind.google/blog/accelerating-fusion-science-through-learned-plasma-control/" ], "title": "Accelerating fusion science through learned plasma control", "organization": "Google DeepMind and the Swiss Plasma Center at EPFL", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2022-02-16", "dateEvidence": "February 16, 2022", "retrievedAt": "2026-09-05T00:33:34.606Z", "reuse": [ { "family": "ai-assisted-fusion-plasma-control", "domains": [ "technology" ], "ids": [ "2037-1" ] } ], "identity": "5f71a37f2bdd52c7ce14489c4b11057ecf3a6f9988a594e350145a1b0d05e35f", "reviewSha256": "1cfd2feb5326e7f300fa02df1d2ff6178a1f843182091f0ef8360a1e1bcfe2f7", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:46.257Z", "lastVerifiedAt": "2026-09-26T21:04:46.257Z", "textSha256": "2f1efd3aae01449b5b94498a839da5f8800a3308c927f1fb7a5bcf5f0cc2a608", "reviewSha256": "1cfd2feb5326e7f300fa02df1d2ff6178a1f843182091f0ef8360a1e1bcfe2f7", "error": null, "etag": "\"GsuqZA\"", "lastModified": null, "retryAt": null } }, "ref-7a6cee9db61a": { "url": "https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/", "urls": [ "https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/" ], "title": "Exploring a space-based, scalable AI infrastructure system design", "organization": "Google Research", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-11-04", "dateEvidence": "November 4, 2025", "retrievedAt": "2026-09-05T00:33:33.928Z", "reuse": [ { "family": "orbital-ai-accelerator-radiation-feasibility", "domains": [ "technology" ], "ids": [ "2039-4" ] } ], "identity": "f8259c3b7762b8017d49a736e5f48282b7dfac86050511007242864a4922f31d", "reviewSha256": "ab6660a32818d3d62fa1846a1c28354198163ee64f681e6a336a12501c6a1fb8", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:46.557Z", "lastVerifiedAt": "2026-09-26T21:04:46.557Z", "textSha256": "c9234ce772eea67302721924d266c5e647c14de096ba19f63ebd8220b3bd5f9a", "reviewSha256": "ab6660a32818d3d62fa1846a1c28354198163ee64f681e6a336a12501c6a1fb8", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-1eba12a4a8fa": { "url": "https://www.anthropic.com/engineering/building-c-compiler", "urls": [ "https://www.anthropic.com/engineering/building-c-compiler" ], "title": "Building a C compiler with a team of parallel Claudes", "organization": "Anthropic", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2026-02-05", "dateEvidence": "Feb 05, 2026", "retrievedAt": "2026-09-05T00:33:41.120Z", "reuse": [ { "family": "long-running-agent-compiler-construction", "domains": [ "technology" ], "ids": [ "2026-0", "2027-1" ] } ], "identity": "e5a0da787dac2cc5ba2aa9f95ab4e1d436fb4388415d42fb5cdc7e9d6f26a26d", "reviewSha256": "1736e0b652863ef45ae103dfdaf4165a92d578c9f9ac39d2e051b68bd214923c", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:47.044Z", "lastVerifiedAt": "2026-09-26T21:04:47.044Z", "textSha256": "e1ca1fb511088306eda14c664a5ba1b0c3891c9df42e4f108b7993f840e0c10e", "reviewSha256": "1736e0b652863ef45ae103dfdaf4165a92d578c9f9ac39d2e051b68bd214923c", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-b05d2a6ff489": { "url": "https://www.anthropic.com/research/how-ai-is-transforming-work-at-anthropic", "urls": [ "https://www.anthropic.com/research/how-ai-is-transforming-work-at-anthropic" ], "title": "How AI Is Transforming Work at Anthropic", "organization": "Anthropic", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-12-02", "dateEvidence": "Dec 2, 2025", "retrievedAt": "2026-09-05T00:33:41.154Z", "reuse": [ { "family": "internal-lab-ai-work-supervision", "domains": [ "technology" ], "ids": [ "2026-1" ] } ], "identity": "6daeff28c3a0f91e9630dd20fd20775b586ec9eef426fbb32df2772696fa6433", "reviewSha256": "cbf764f57fb8d5538ebb67e76146626b127eb601cfaa6a4053aede1148d3952e", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:47.748Z", "lastVerifiedAt": "2026-09-26T21:04:47.748Z", "textSha256": "5c1df3b9a0af00aafc1a563d67687fabf36e8750b4da04256e921e3115c73830", "reviewSha256": "cbf764f57fb8d5538ebb67e76146626b127eb601cfaa6a4053aede1148d3952e", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-05665ed86ec7": { "url": "https://lastexam.ai/", "urls": [ "https://lastexam.ai/" ], "title": "Humanity's Last Exam", "organization": "Center for AI Safety, Scale AI, and HLE Contributors Consortium", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:33:41.580Z", "reuse": [ { "family": "cross-domain-closed-ended-expert-evaluation", "domains": [ "technology" ], "ids": [ "2026-6", "2035-0" ] } ], "identity": "e87e7cefd5df2fc308481027322b9e4c52ff4d76c38ce609049c5c0d884a76a0", "reviewSha256": "158f2094f0ff904713fcb95e123131b4c7bd93c2e87206a13c3a39e2c01be168", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:48.948Z", "lastVerifiedAt": "2026-09-26T21:04:48.948Z", "textSha256": "1487bbbf202d33408eca1db3da698625ace0666fb99dd50087ef2bb915e1e993", "reviewSha256": "158f2094f0ff904713fcb95e123131b4c7bd93c2e87206a13c3a39e2c01be168", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-056e0ff33675": { "url": "https://metr.org/blog/2024-11-22-evaluating-r-d-capabilities-of-llms/", "urls": [ "https://metr.org/blog/2024-11-22-evaluating-r-d-capabilities-of-llms/" ], "title": "Evaluating frontier AI R&D capabilities of language model agents against human experts", "organization": "METR", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2024-11-22", "dateEvidence": "2024/11/22", "retrievedAt": "2026-09-05T00:33:41.677Z", "reuse": [ { "family": "bounded-ml-research-engineering-evaluation", "domains": [ "technology" ], "ids": [ "2027-2", "2030-0", "2031-0" ] } ], "identity": "84043690af693ef3a3b6cad9b38ec30ba847357ca60657dc995926bc41f198ad", "reviewSha256": "d66aae290314a145dcb326947f483ed6e6f5253adda2fe6766024d8dce733823", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:49.475Z", "lastVerifiedAt": "2026-09-25T21:04:41.691Z", "textSha256": "9924cc1756fa8b5b127d1abc36fd795a1267e155dfe1bf25736a3acd3757c440", "reviewSha256": "d66aae290314a145dcb326947f483ed6e6f5253adda2fe6766024d8dce733823", "error": null, "etag": "\"f8fe8448f381dc9b752064a00ccae5ba-ssl-df\"", "lastModified": null, "retryAt": null } }, "ref-d05d86b6fe3b": { "url": "https://openai.com/index/gdpval/", "urls": [ "https://openai.com/index/gdpval/" ], "title": "Measuring the performance of our models on real-world tasks", "organization": "OpenAI", "quality": "official-research", "transport": "browser", "format": "html", "publishedAt": "2025-09-25", "dateEvidence": "September 25, 2025", "retrievedAt": "2026-09-05T00:34:41.188Z", "reuse": [ { "family": "expert-curated-occupational-evaluation", "domains": [ "technology" ], "ids": [ "2028-2" ] } ], "identity": "358a5a312da8b91f196bf25b9c06682143f2e9bef8feb8d97fd081b9856b3d67", "reviewSha256": "3a92704b4b2beb6615711146eac7f58450dd2ff23269e8b643f24e3831dc81a4", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:53.016Z", "lastVerifiedAt": "2026-09-26T21:04:53.016Z", "textSha256": "25773bd39ac56806e040abdaf739e31cab975a71e1d79ae95b96dc2c4474073a", "reviewSha256": "3a92704b4b2beb6615711146eac7f58450dd2ff23269e8b643f24e3831dc81a4", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-1b5c7b499756": { "url": "https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/", "urls": [ "https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/" ], "title": "AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms", "organization": "Google DeepMind", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-05-14", "dateEvidence": "2025-05-14T00:00:00+00:00", "retrievedAt": "2026-09-05T00:33:41.955Z", "reuse": [ { "family": "ai-generated-training-kernel-feedback", "domains": [ "technology" ], "ids": [ "2028-5" ] } ], "identity": "096db4c551319fcfcccab5c36407d98c8cf70dfe871cc642008274642b7f1ef1", "reviewSha256": "42185ce779f2fbce76d6ea60f5a702c9994686d12672346f4fb978977f7239b4", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:54.358Z", "lastVerifiedAt": "2026-09-26T21:04:54.358Z", "textSha256": "129ed2c4b8db89839562bc52e4f1d716e4d136fcd5f6cb78689cb6e78550810a", "reviewSha256": "42185ce779f2fbce76d6ea60f5a702c9994686d12672346f4fb978977f7239b4", "error": null, "etag": "\"GsuqZA\"", "lastModified": null, "retryAt": null } }, "ref-443d9796ce11": { "url": "https://developer.nvidia.com/blog/confidential-computing-on-h100-gpus-for-secure-and-trustworthy-ai/", "urls": [ "https://developer.nvidia.com/blog/confidential-computing-on-h100-gpus-for-secure-and-trustworthy-ai/" ], "title": "Confidential Computing on NVIDIA H100 GPUs for Secure and Trustworthy AI", "organization": "NVIDIA", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2023-08-03", "dateEvidence": "2023-08-03T22:32:37+00:00", "retrievedAt": "2026-09-05T00:33:42.786Z", "reuse": [ { "family": "accelerator-state-attestation-prerequisite", "domains": [ "technology" ], "ids": [ "2029-3" ] } ], "identity": "fe4fb64d1ccc959cd1f735e3b47e0033819164fd982a50475fc96b52b303a05e", "reviewSha256": "b978f418241e44b9a650f67446faebe1dd9e11290e6a4876426ca22b4201444f", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:56.216Z", "lastVerifiedAt": "2026-09-26T21:04:56.216Z", "textSha256": "c24dce52439a8174a4b4e8e1fbd7dc803d0a6c948be135b96d44896575ecb6bd", "reviewSha256": "b978f418241e44b9a650f67446faebe1dd9e11290e6a4876426ca22b4201444f", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-1f65760bea2b": { "url": "https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/", "urls": [ "https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/" ], "title": "Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad", "organization": "Google DeepMind", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-07-21", "dateEvidence": "2025-07-21T00:00:00+00:00", "retrievedAt": "2026-09-05T00:33:41.956Z", "reuse": [ { "family": "externally-graded-olympiad-reasoning", "domains": [ "technology" ], "ids": [ "2030-1" ] } ], "identity": "4e5a6b1d99d21d3fe763b42e1f5e4142268e5c0933a9cc0275cc486aec69e049", "reviewSha256": "6842bfe6a79b756fad7cee32f6e546b35c7be3c0001088e85d6dffcf5f355fe6", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:57.497Z", "lastVerifiedAt": "2026-09-26T21:04:57.497Z", "textSha256": "603fd055302bed0eb65572e1308b2dbb7f30e3a06d6b01941f254fb1989f3252", "reviewSha256": "6842bfe6a79b756fad7cee32f6e546b35c7be3c0001088e85d6dffcf5f355fe6", "error": null, "etag": "\"GsuqZA\"", "lastModified": null, "retryAt": null } }, "ref-58a0c016acd5": { "url": "https://research.google/blog/titans-miras-helping-ai-have-long-term-memory/", "urls": [ "https://research.google/blog/titans-miras-helping-ai-have-long-term-memory/" ], "title": "Titans + MIRAS: Helping AI have long-term memory", "organization": "Google Research", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-12-04", "dateEvidence": "data-blog-publish-date=\"20251204\"", "retrievedAt": "2026-09-05T00:33:41.520Z", "reuse": [ { "family": "test-time-parameterized-memory-learning", "domains": [ "technology" ], "ids": [ "2031-5" ] } ], "identity": "a257beb5451518ccec06a9204363c2819e09a1aac6ee5b8a79899c52f70cbd22", "reviewSha256": "c97c50567868a27023b2dec442cd6ea8782ecb95ed34e81c74e72863ff0832cb", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:57.790Z", "lastVerifiedAt": "2026-09-26T21:04:57.790Z", "textSha256": "b69f7e061d4d4b1138255d939181b37ce411bbaf0bd84a9516658437b089d9e6", "reviewSha256": "c97c50567868a27023b2dec442cd6ea8782ecb95ed34e81c74e72863ff0832cb", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-fdc10e3226e6": { "url": "https://www.anthropic.com/research/tracing-thoughts-language-model", "urls": [ "https://www.anthropic.com/research/tracing-thoughts-language-model" ], "title": "Tracing the thoughts of a large language model", "organization": "Anthropic", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-03-27", "dateEvidence": "Mar 27, 2025", "retrievedAt": "2026-09-05T00:33:41.110Z", "reuse": [ { "family": "attribution-circuit-reasoning-faithfulness", "domains": [ "technology" ], "ids": [ "2035-6", "2038-1" ] } ], "identity": "5509451faeb4158540ad90807e2aa684ac9a10934d6d10042487c40726dcb030", "reviewSha256": "d3fea08afcd3477ba23c4db7ab0e142683c59cd0ab4c09e182e49e629fca53cc", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:58.282Z", "lastVerifiedAt": "2026-09-26T21:04:58.282Z", "textSha256": "2f689bd81cb0151e6956f37e6428bfe878d3c23043fe43996a5fd4f124fa446e", "reviewSha256": "d3fea08afcd3477ba23c4db7ab0e142683c59cd0ab4c09e182e49e629fca53cc", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-a130b8e6da20": { "url": "https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/", "urls": [ "https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/" ], "title": "Accelerating scientific breakthroughs with an AI co-scientist", "organization": "Google Research and collaborating biomedical research institutions", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-02-19", "dateEvidence": "data-blog-publish-date=\"20250219\"", "retrievedAt": "2026-09-05T00:33:41.529Z", "reuse": [ { "family": "ai-hypothesis-generation-with-laboratory-validation", "domains": [ "technology" ], "ids": [ "2037-0" ] } ], "identity": "5434dee876797a93ea95bc78f1c2265405ab99023455de6d8523b66d92d933d3", "reviewSha256": "6ad6eb813d70900f28a39f78ccb0b48860b4a7034517c45564bec23e97239599", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:58.559Z", "lastVerifiedAt": "2026-09-26T21:04:58.559Z", "textSha256": "646338b91548307547f9ff65e2a83a3253d0793cc03386d96c5ce6b6cad29875", "reviewSha256": "6ad6eb813d70900f28a39f78ccb0b48860b4a7034517c45564bec23e97239599", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-c2cfd72baafd": { "url": "https://www.anthropic.com/research/alignment-faking", "urls": [ "https://www.anthropic.com/research/alignment-faking" ], "title": "Alignment faking in large language models", "organization": "Anthropic and Redwood Research", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2024-12-18", "dateEvidence": "Dec 18, 2024", "retrievedAt": "2026-09-05T00:33:41.155Z", "reuse": [ { "family": "experimental-preference-preservation-under-training", "domains": [ "technology" ], "ids": [ "2038-0" ] } ], "identity": "1ee8ed56e85822b2e18c6d9d7469c8f6d6eb36f07299d18f5ba09f8d9edd8258", "reviewSha256": "67dcd02110906ac0fd392d398b5be5fe259f854d085376ba9e6b8b5aa0735c3e", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:59.089Z", "lastVerifiedAt": "2026-09-26T21:04:59.089Z", "textSha256": "7fab1f53349f224378534487d402edb43fdbd758df1c6acfb265a0d9bc5e2c26", "reviewSha256": "67dcd02110906ac0fd392d398b5be5fe259f854d085376ba9e6b8b5aa0735c3e", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-83b187f91a7c": { "url": "https://www.anthropic.com/research/sleeper-agents-training-deceptive-llms-that-persist-through-safety-training", "urls": [ "https://www.anthropic.com/research/sleeper-agents-training-deceptive-llms-that-persist-through-safety-training" ], "title": "Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training", "organization": "Anthropic and research collaborators", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2024-01-14", "dateEvidence": "Jan 14, 2024", "retrievedAt": "2026-09-05T00:33:41.091Z", "reuse": [ { "family": "safety-training-resistant-conditional-behavior", "domains": [ "technology" ], "ids": [ "2038-2" ] } ], "identity": "b811e66cebf6c97fbca0153bfd2fc826cc8f52e5b82f5c78fff19f13fda4cec0", "reviewSha256": "7bb7bd105738f13aa80528db2a9ec0ef9013d82e2a16d1ac895c167cc4ae5a82", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:59.441Z", "lastVerifiedAt": "2026-09-26T21:04:59.441Z", "textSha256": "97714528a5406f5528bdaa69f76c73b87da2f5f94c866fb67c2ba94c8d3ca2a6", "reviewSha256": "7bb7bd105738f13aa80528db2a9ec0ef9013d82e2a16d1ac895c167cc4ae5a82", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-0b707017d000": { "url": "https://www.anthropic.com/research/auditing-hidden-objectives", "urls": [ "https://www.anthropic.com/research/auditing-hidden-objectives" ], "title": "Auditing language models for hidden objectives", "organization": "Anthropic", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-03-13", "dateEvidence": "Mar 13, 2025", "retrievedAt": "2026-09-05T00:33:41.403Z", "reuse": [ { "family": "blinded-hidden-objective-alignment-auditing", "domains": [ "technology" ], "ids": [ "2039-1" ] } ], "identity": "f69239661f4893833c6bedf12f1b5040ccff07f7f971c20e985310682db011e7", "reviewSha256": "b846abacd1755d15d5a226e8e684e65caa2b877fcc21f68ce7dda1c7db8c617e", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:04:59.918Z", "lastVerifiedAt": "2026-09-26T21:04:59.918Z", "textSha256": "295c1ae7b2c014c04759a66299fed43314695c29959db5d53f6d35e810451975", "reviewSha256": "b846abacd1755d15d5a226e8e684e65caa2b877fcc21f68ce7dda1c7db8c617e", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-d485fa5001db": { "url": "https://group.softbank/en/news/press/20260227", "urls": [ "https://group.softbank/en/news/press/20260227" ], "title": "Follow-on Investments in OpenAI", "organization": "SoftBank Group Corp.", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2026-02-27", "dateEvidence": "February 27, 2026", "retrievedAt": "2026-09-05T00:36:27.6305886Z", "reuse": [ { "family": "frontier-lab-contractual-equity-valuation", "domains": [ "economic" ], "ids": [ "2026-2" ] } ], "identity": "ab5875deb39a0c3b43ce12cae57c8991590af3ae29b79dea1db9e058be44db55", "reviewSha256": "2a83994555f0776498aa01714a46cc4fde6d5303afb1274f80f0604bffa001f9", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:00.666Z", "lastVerifiedAt": "2026-09-25T21:04:53.675Z", "textSha256": "da0f66ca27c864c533dc42c8f5e205c697d2de4ed9441223b12c5d70c9fb2157", "reviewSha256": "2a83994555f0776498aa01714a46cc4fde6d5303afb1274f80f0604bffa001f9", "error": null, "etag": "W/\"11x22h69z6q4f80\"", "lastModified": null, "retryAt": null } }, "ref-a06d967a4287": { "url": "https://www.salesforce.com/news/press-releases/2026/02/25/fy26-q4-earnings/", "urls": [ "https://www.salesforce.com/news/press-releases/2026/02/25/fy26-q4-earnings/" ], "title": "Salesforce Announces Fourth Quarter and Full Fiscal 2026 Results", "organization": "Salesforce", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2026-02-25", "dateEvidence": "February 25, 2026", "retrievedAt": "2026-09-05T00:36:27.6305884Z", "reuse": [ { "family": "salesforce-commercial-agent-work-deployment", "domains": [ "economic" ], "ids": [ "2027-0", "2034-1" ] } ], "identity": "dfa3bc89c122af21bd5acd3a870d8bf51f777d6d1621d2dbb3148d1405822128", "reviewSha256": "a48599d46b6c61db850bfe2347f5d632e10e100ec68187867067ce9865107977", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:01.329Z", "lastVerifiedAt": "2026-09-26T21:05:01.329Z", "textSha256": "8388bd88596c7b4fcbbfc3b555db27817200ab000f6390eae5242966b53730c7", "reviewSha256": "a48599d46b6c61db850bfe2347f5d632e10e100ec68187867067ce9865107977", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-2a44176650f8": { "url": "https://www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-occupational-exposure", "urls": [ "https://www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-occupational-exposure" ], "title": "Generative AI and Jobs: A Refined Global Index of Occupational Exposure", "organization": "International Labour Organization; Pawel Gmyrek and coauthors", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2025-05-20", "dateEvidence": "20 May 2025", "retrievedAt": "2026-09-05T00:36:27.6307527Z", "reuse": [ { "family": "occupation-task-exposure-and-human-complements", "domains": [ "economic" ], "ids": [ "2027-3" ] } ], "identity": "62eaf00cacb9850a991cc7ae479ba5b3041aaac8fa4e2c393bbd3446e29f64f3", "reviewSha256": "74913fdb9b1fd445356638efc50bbb20c55e0a94d345dcf055a8397ceb8f1115", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:02.246Z", "lastVerifiedAt": "2026-09-26T21:05:02.246Z", "textSha256": "d450cdfc107a47f35c04f70db66f6a912ae801e5a65f29058248ee5ca91630ee", "reviewSha256": "74913fdb9b1fd445356638efc50bbb20c55e0a94d345dcf055a8397ceb8f1115", "error": null, "etag": null, "lastModified": "Fri, 25 Sep 2026 14:28:27 GMT", "retryAt": null } }, "ref-92735c509f66": { "url": "https://www.nber.org/papers/w31161", "urls": [ "https://www.nber.org/papers/w31161" ], "title": "Generative AI at Work", "organization": "Erik Brynjolfsson, Danielle Li and Lindsey R. Raymond; NBER working paper", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": "April 2023", "retrievedAt": "2026-09-05T00:36:27.6559718Z", "reuse": [ { "family": "customer-support-human-ai-workflow-productivity", "domains": [ "economic" ], "ids": [ "2028-1" ] } ], "publishedPeriod": "2023-04", "identity": "87946e951684614850867cda45c661ded589299194fec4efe7d4493151c04202", "reviewSha256": "2591ce26634f62c8bf99d73e473e4c55a813b000960350c3a6225155259a9440", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:02.846Z", "lastVerifiedAt": "2026-09-26T21:05:02.846Z", "textSha256": "59ad01f0a2e9555fd08efa3132078a407ee9c267a0f8eff5b6bfe64e2ac00c7f", "reviewSha256": "2591ce26634f62c8bf99d73e473e4c55a813b000960350c3a6225155259a9440", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-52dbf968dc6c": { "url": "https://www.stlouisfed.org/on-the-economy/2024/sep/rapid-adoption-generative-ai", "urls": [ "https://www.stlouisfed.org/on-the-economy/2024/sep/rapid-adoption-generative-ai" ], "title": "The Rapid Adoption of Generative AI", "organization": "Alexander Bick, Adam Blandin and David Deming; Federal Reserve Bank of St. Louis", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2024-09-23", "dateEvidence": "9/23/2024 4:30 AM", "retrievedAt": "2026-09-05T00:37:15.9196100Z", "reuse": [ { "family": "survey-measured-ai-assisted-work-hour-intensity", "domains": [ "economic" ], "ids": [ "2029-0" ] } ], "identity": "ecf9f0e1281decaddea73aba240ef27af79d24ff93f72319816f3f7171e4e6ff", "reviewSha256": "446621c7691b42ebec03220601609e66d0f42829126f531691eecc6a2a89b93b", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:04.372Z", "lastVerifiedAt": "2026-09-26T21:05:04.372Z", "textSha256": "a21ac9135264560a998f458be0c9ee15d08365fa43129e6aa142ad3388116789", "reviewSha256": "446621c7691b42ebec03220601609e66d0f42829126f531691eecc6a2a89b93b", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-aab8dd5402ac": { "url": "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2026", "urls": [ "https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2026" ], "title": "NVIDIA Announces Financial Results for First Quarter Fiscal 2026", "organization": "NVIDIA", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2025-05-28", "dateEvidence": "May 28, 2025", "retrievedAt": "2026-09-05T00:36:27.6634022Z", "reuse": [ { "family": "ai-chip-export-control-earnings-shock", "domains": [ "social" ], "ids": [ "2029-6" ] } ], "identity": "629d396513819dcbfd447251d6dc864e31c36458c4c765506d42e4f0590b9dab", "reviewSha256": "512de728b647365467cf97606139508563c4805568317e72287085e45a83c0a4", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:05.375Z", "lastVerifiedAt": "2026-09-26T21:05:05.375Z", "textSha256": "303fae8f2239afeb6cc0ce0bacf60dda9395de83fc820420aa60877cc7542b28", "reviewSha256": "512de728b647365467cf97606139508563c4805568317e72287085e45a83c0a4", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-9c5b9fb39a11": { "url": "https://ifr.org/ifr-press-releases/news/robot-density-surges-in-europe-asia-and-americas", "urls": [ "https://ifr.org/ifr-press-releases/news/robot-density-surges-in-europe-asia-and-americas" ], "title": "Robot Density Surges in Europe, Asia, and Americas", "organization": "International Federation of Robotics", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2026-04-08", "dateEvidence": "Frankfurt, Apr 08, 2026", "retrievedAt": "2026-09-07T02:08:26.947Z", "reuse": [ { "family": "manufacturing-industrial-robot-installed-density", "domains": [ "economic" ], "ids": [ "2031-1" ] } ], "identity": "b86a60cb66b151ca514ec6348c73744ce5a8ae7b90911ebe605465462287fff2", "reviewSha256": "4abf8a36767504d8692a1a1c36cd6e6277d28fb81ec8944e843ee8f4e1f7bd66", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:05.783Z", "lastVerifiedAt": "2026-09-26T21:05:05.783Z", "textSha256": "e7ef4a796ea6d406b4fcb0c4df60a3fb54e8906ed0fe7752643540c5d146812f", "reviewSha256": "4abf8a36767504d8692a1a1c36cd6e6277d28fb81ec8944e843ee8f4e1f7bd66", "error": null, "etag": null, "lastModified": "Sat, 26 Sep 2026 21:05:06 GMT", "retryAt": null } }, "ref-a0af5ab6537d": { "url": "https://www.microsoft.com/en-us/Investor/earnings/FY-2026-Q4/press-release-webcast", "urls": [ "https://www.microsoft.com/en-us/Investor/earnings/FY-2026-Q4/press-release-webcast" ], "title": "Microsoft Cloud and AI Strength Fuels Fourth Quarter Results", "organization": "Microsoft", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2026-07-29", "dateEvidence": "July 29, 2026", "retrievedAt": "2026-09-05T01:02:28.7418644Z", "reuse": [ { "family": "ai-platform-company-reported-annual-revenue", "domains": [ "economic" ], "ids": [ "2031-2" ] } ], "identity": "e9866d3dbb7dcb12ca3d2456aababbda49cce229adae97c490e2e2e5b0b61141", "reviewSha256": "b7bc30137c551ba0b28f01e70590b61ee53322c74435624e9676fa00109639ac", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:06.050Z", "lastVerifiedAt": "2026-09-26T21:05:06.050Z", "textSha256": "5656221f9b20145800ab1947ac9ef197412ff33db39e5d997590584a84b281f3", "reviewSha256": "b7bc30137c551ba0b28f01e70590b61ee53322c74435624e9676fa00109639ac", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-2be84f63d7cb": { "url": "https://www.anthropic.com/news/the-anthropic-economic-index", "urls": [ "https://www.anthropic.com/news/the-anthropic-economic-index" ], "title": "The Anthropic Economic Index", "organization": "Anthropic", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2025-02-10", "dateEvidence": "Feb 10, 2025", "retrievedAt": "2026-09-05T00:36:28.3852486Z", "reuse": [ { "family": "observed-llm-occupational-task-coverage", "domains": [ "economic" ], "ids": [ "2032-0" ] } ], "identity": "c5d45bbcb83c2076f5bc9d070748a961120792ed50b2b51c27685f835e27ff47", "reviewSha256": "d2bc7de35704cbd6400936e5edecc27774fcb83ba79101ac06d9e8847cf2fa3d", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:06.597Z", "lastVerifiedAt": "2026-09-26T21:05:06.597Z", "textSha256": "0b1f4e4c1911b417465505a443ee52135b71448622e32641979e4c8f4928be87", "reviewSha256": "d2bc7de35704cbd6400936e5edecc27774fcb83ba79101ac06d9e8847cf2fa3d", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-569faca09ce6": { "url": "https://www.nber.org/papers/w32487", "urls": [ "https://www.nber.org/papers/w32487" ], "title": "The Simple Macroeconomics of AI", "organization": "Daron Acemoglu; NBER working paper", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": "May 2024", "retrievedAt": "2026-09-05T00:36:28.7298062Z", "reuse": [ { "family": "task-cost-savings-to-aggregate-productivity-constraint", "domains": [ "economic" ], "ids": [ "2032-2" ] } ], "publishedPeriod": "2024-05", "identity": "a93c581d0eb25cdd94fdc50e1e19ee7100b7d9a29ae34317ea59b040f4d6b32e", "reviewSha256": "a03fbc4afd49a45149981132ed58758e8522b1912fa5fc0268691b6942b3b1eb", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:06.929Z", "lastVerifiedAt": "2026-09-26T21:05:06.929Z", "textSha256": "ae7314b485050325a305a36023607ab52bac622169745d86afc64923353f8f07", "reviewSha256": "a03fbc4afd49a45149981132ed58758e8522b1912fa5fc0268691b6942b3b1eb", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-57bd7c78c8c4": { "url": "https://www.iza.org/publications/dp/8938/robots-at-work", "urls": [ "https://www.iza.org/publications/dp/8938/robots-at-work" ], "title": "Robots at Work", "organization": "Georg Graetz and Guy Michaels; IZA discussion paper", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": "March 2015", "retrievedAt": "2026-09-05T00:36:28.8575439Z", "reuse": [ { "family": "industrial-robot-value-added-growth-contribution", "domains": [ "economic" ], "ids": [ "2033-0" ] } ], "publishedPeriod": "2015-03", "identity": "70ca61bf9372f0baaac34647469ab236d6299cd83b6983a080ac46cb1640906a", "reviewSha256": "ebc8542cb109643727cd8f08f7bda7e7146de4c3837dcf8afd6f1faf400e06aa", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:08.599Z", "lastVerifiedAt": "2026-09-26T21:05:08.599Z", "textSha256": "eac26bd685f0a389a721f50c13d3182fc964ed79e4354030b7800344155224e9", "reviewSha256": "ebc8542cb109643727cd8f08f7bda7e7146de4c3837dcf8afd6f1faf400e06aa", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-1b81e5c02a10": { "url": "https://apfc.org/the-fund/fund-structure/", "urls": [ "https://apfc.org/the-fund/fund-structure/" ], "title": "Fund Structure", "organization": "Alaska Permanent Fund Corporation, State of Alaska", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:36:29.1033617Z", "reuse": [ { "family": "alaska-rent-funded-statutory-resident-dividend", "domains": [ "economic" ], "ids": [ "2033-1", "2035-3" ] } ], "identity": "72c60dd769709001646794c4df246edf028f1fdfee3bae4892b2b8384f15e108", "reviewSha256": "217ad2ee568dcde07ed69df7234bbb8da07a628b094ff42843f8262d414b91a8", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:09.108Z", "lastVerifiedAt": "2026-09-26T21:05:09.108Z", "textSha256": "abd0e4d2f60fca3330334f2f263572a9943821d28876c653c88c34445027a1a9", "reviewSha256": "217ad2ee568dcde07ed69df7234bbb8da07a628b094ff42843f8262d414b91a8", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-5007e8060d43": { "url": "https://blog.google/products-and-platforms/products/search/ai-mode-search/", "urls": [ "https://blog.google/products-and-platforms/products/search/ai-mode-search/" ], "title": "Expanding AI Overviews and introducing AI Mode", "organization": "Google; Robby Stein, VP of Product, Google Search", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2025-03-05", "dateEvidence": "Mar 05, 2025", "retrievedAt": "2026-09-05T00:36:29.2546565Z", "reuse": [ { "family": "conversational-source-linked-search-interface-trial", "domains": [ "individual" ], "ids": [ "2033-4" ] } ], "identity": "5f33cda2761d6db55047a21c6ee678d996f341aaaade82d7069c04d07c3638d2", "reviewSha256": "142a3085b9b7557200540cf7bdcb566423ce8c64b19ac91bcf3676a4ca4730cf", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:09.692Z", "lastVerifiedAt": "2026-09-26T21:05:09.692Z", "textSha256": "f71790d9fe1e9235d80f07e35e2742f863d70b4519e4d51771a956022b0043e9", "reviewSha256": "142a3085b9b7557200540cf7bdcb566423ce8c64b19ac91bcf3676a4ca4730cf", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-ee87371859e6": { "url": "https://www.nber.org/papers/w23928", "urls": [ "https://www.nber.org/papers/w23928" ], "title": "Artificial Intelligence and Economic Growth", "organization": "Philippe Aghion, Benjamin F. Jones and Charles I. Jones; NBER working paper", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": "October 2017", "retrievedAt": "2026-09-05T00:36:29.2701059Z", "reuse": [ { "family": "essential-input-bottlenecks-in-ai-macro-growth", "domains": [ "economic" ], "ids": [ "2033-5", "2036-3" ] } ], "publishedPeriod": "2017-10", "identity": "2e3ccf9a1a18340d4a1ce4cedb6ca6ce0e68559c6e87d93da0f6c01689041353", "reviewSha256": "6180e8a01821773008abc3923eeaf577b85446edc5d46cbd6fba653f5618db7d", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:10.275Z", "lastVerifiedAt": "2026-09-26T21:05:10.275Z", "textSha256": "4bc8d3fabcd5ddb8c6630375f65d9f44e59e06511c6b864cc91c55a1e8aa8c91", "reviewSha256": "6180e8a01821773008abc3923eeaf577b85446edc5d46cbd6fba653f5618db7d", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-d9afef730d39": { "url": "https://www.nber.org/papers/w35271", "urls": [ "https://www.nber.org/papers/w35271" ], "title": "Beyond Exposure: Predicting AI Adoption Based on Comparative Advantage", "organization": "Ilse Lindenlaub, Ryungha Oh, Maria Alejandra Rodriguez and Laura Veldkamp via NBER", "quality": "author-manuscript", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": "May 2026", "retrievedAt": "2026-09-05T09:53:43.793Z", "reuse": [ { "family": "comparative-advantage-ai-adoption-economics", "domains": [ "economic" ], "ids": [ "2035-2" ] } ], "publishedPeriod": "2026-05", "identity": "3849b2bfdd541ad4f81f3997a058a4cee81cf50227d3fed82c30a5fb0a54f1da", "reviewSha256": "855eb25bce898ab10a3711da82642dbb170ed4938e3f290de7ef0c6e28177ba0", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:11.279Z", "lastVerifiedAt": "2026-09-26T21:05:11.279Z", "textSha256": "84f53979e0ef0a52f01b9c306131fbbf366dd2aea51c1c43d2d7e0d7ca4245b3", "reviewSha256": "855eb25bce898ab10a3711da82642dbb170ed4938e3f290de7ef0c6e28177ba0", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-0e4704ac5d99": { "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12367540/", "urls": [ "https://pmc.ncbi.nlm.nih.gov/articles/PMC12367540/" ], "title": "On the conversational persuasiveness of GPT-4", "organization": "Francesco Salvi, Manoel Horta Ribeiro, Riccardo Gallotti and Robert West; EPFL, Fondazione Bruno Kessler and Princeton University; Nature Human Behaviour", "quality": "peer-reviewed", "transport": "https", "format": "html", "publishedAt": "2025-05-19", "dateEvidence": "2025 May 19", "retrievedAt": "2026-09-05T00:36:29.9518764Z", "reuse": [ { "family": "controlled-human-versus-llm-conversational-persuasion", "domains": [ "social" ], "ids": [ "2035-4" ] } ], "identity": "ac132247520e43069ebac81a43061f5703cda10658e5cbb2b8d63934b2469d29", "reviewSha256": "8812605108358b770e60b92453494d9a6dad2a6d5f775b4535e890d6cadeea2e", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:11.801Z", "lastVerifiedAt": "2026-09-26T21:05:11.801Z", "textSha256": "34f385e8dc1f6cf16bdfe388accdf017b2f2126cd0e95f3995b6643634917142", "reviewSha256": "8812605108358b770e60b92453494d9a6dad2a6d5f775b4535e890d6cadeea2e", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-5cea71b71d6e": { "url": "https://www.anthropic.com/research/exploring-model-welfare", "urls": [ "https://www.anthropic.com/research/exploring-model-welfare" ], "title": "Exploring model welfare", "organization": "Anthropic", "quality": "official-company", "transport": "https", "format": "html", "publishedAt": "2025-04-24", "dateEvidence": "Apr 24, 2025", "retrievedAt": "2026-09-05T00:36:29.9578165Z", "reuse": [ { "family": "corporate-model-welfare-research-institutionalization", "domains": [ "social" ], "ids": [ "2035-5" ] } ], "identity": "7012b309e5e5cc4d40983447c851b479bf05b6ca141b6bb46a9a37578cb018f3", "reviewSha256": "a220716a403bf74066c24f97eaae32233c03a6d18f16dc4be17f50a6e7679eff", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:12.464Z", "lastVerifiedAt": "2026-09-26T21:05:12.464Z", "textSha256": "b97ca24514377f85756f18332cbff4d436e4edcc496e46b8e1a0fa8539f3f376", "reviewSha256": "a220716a403bf74066c24f97eaae32233c03a6d18f16dc4be17f50a6e7679eff", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-5efeb9ec3c3d": { "url": "https://www.ons.gov.uk/employmentandlabourmarket/peopleinwork/employmentandemployeetypes/bulletins/employmentintheuk/august2026", "urls": [ "https://www.ons.gov.uk/employmentandlabourmarket/peopleinwork/employmentandemployeetypes/bulletins/employmentintheuk/august2026" ], "title": "Employment in the UK: August 2026", "organization": "UK Office for National Statistics", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2026-08-18", "dateEvidence": "18 August 2026", "retrievedAt": "2026-09-05T01:02:29.0671651Z", "reuse": [ { "family": "working-age-employment-population-baseline", "domains": [ "economic" ], "ids": [ "2036-2" ] } ], "identity": "34a080f031e1b38c48eba871ee255013e10e90620e8e59e7e14bc1b28c227b98", "reviewSha256": "45fc3985f3338a4264f2f7a06bac6826982231c13050c619bea348569fe2ca2a", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:12.684Z", "lastVerifiedAt": "2026-09-26T21:05:12.684Z", "textSha256": "44384088bdf8d91ea369b4ad32d59e63c68572236c16c87f4624d818df48bfef", "reviewSha256": "45fc3985f3338a4264f2f7a06bac6826982231c13050c619bea348569fe2ca2a", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-5f6f63c537e3": { "url": "https://www.oecd.org/en/data/tools/oecd-learning-compass-2030.html", "urls": [ "https://www.oecd.org/en/data/tools/oecd-learning-compass-2030.html" ], "title": "The OECD Learning Compass 2030", "organization": "Organisation for Economic Co-operation and Development", "quality": "official-research", "transport": "browser", "format": "html", "publishedAt": "2024-06-20T14:20:00Z", "dateEvidence": "2024-06-20T14:20:00Z", "retrievedAt": "2026-09-05T00:37:15.9196100Z", "reuse": [ { "family": "education-agency-wellbeing-responsibility-framework", "domains": [ "individual" ], "ids": [ "2036-6" ] } ], "identity": "a08b03dcb9f77d012c660547810c85561c69e5bfc0a6a95391635bfd1fa1d505", "reviewSha256": "ee16cd82bbcb514f4ac557f9f9a652e20846ef6e78a91ce2617124fd84c37a3b", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:15.903Z", "lastVerifiedAt": "2026-09-26T21:05:15.903Z", "textSha256": "5f0ee58aefeeadd49642275644aedf27f4b2705bbdecce061a432f76e50427b5", "reviewSha256": "ee16cd82bbcb514f4ac557f9f9a652e20846ef6e78a91ce2617124fd84c37a3b", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-21375cbb36fd": { "url": "https://www.pewresearch.org/science/2023/11/14/americans-trust-in-scientists-positive-views-of-science-continue-to-decline/", "urls": [ "https://www.pewresearch.org/science/2023/11/14/americans-trust-in-scientists-positive-views-of-science-continue-to-decline/" ], "title": "Americans’ Trust in Scientists, Positive Views of Science Continue to Decline", "organization": "Pew Research Center", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:36:30.3670218Z", "reuse": [ { "family": "science-authority-partisan-attitude-change", "domains": [ "social" ], "ids": [ "2037-4" ] } ], "identity": "7a947460b6d6491b1827a1d8f69da3a01f10fffe551ec1fd9a3706715fdd9ec6", "reviewSha256": "bebb230a4dfce4ee4222d2a52e203c465b15f74d15e66324daf0ce924b19afd5", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:16.258Z", "lastVerifiedAt": "2026-09-26T21:05:16.258Z", "textSha256": "1909e16a714009b2c0a08485f5364df4344791b9de50da803e4e631af8af7e70", "reviewSha256": "bebb230a4dfce4ee4222d2a52e203c465b15f74d15e66324daf0ce924b19afd5", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-587a6715a0cb": { "url": "https://www.gov.uk/government/publications/ai-safety-institute-overview/introducing-the-ai-safety-institute", "urls": [ "https://www.gov.uk/government/publications/ai-safety-institute-overview/introducing-the-ai-safety-institute" ], "title": "Introducing the AI Safety Institute", "organization": "UK Department for Science, Innovation and Technology", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": "November 2023", "retrievedAt": "2026-09-05T00:36:30.4482624Z", "reuse": [ { "family": "public-frontier-ai-safety-institution-funding", "domains": [ "economic" ], "ids": [ "2038-4" ] } ], "publishedPeriod": "2023-11", "identity": "f080ea14e89bc7eef6c4e422ffcf6c1ddf535dd54681186ccaa03cffcf21ddaa", "reviewSha256": "77238162e5153a2a0e92a3e86d0174854c31a81e9178fc848b9c21dfde914a32", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:16.732Z", "lastVerifiedAt": "2026-09-26T21:05:16.732Z", "textSha256": "63feba305f71dca548627542bf04e58362fd6dc452a7d5c5abb44e8985347097", "reviewSha256": "77238162e5153a2a0e92a3e86d0174854c31a81e9178fc848b9c21dfde914a32", "error": null, "etag": "W/\"262f2f0c27fd114935699c9ea4fc449c\"", "lastModified": null, "retryAt": null } }, "ref-e09b4949277a": { "url": "https://pfd.alaska.gov/Division-Info/summary-of-dividend-applications-payments", "urls": [ "https://pfd.alaska.gov/Division-Info/summary-of-dividend-applications-payments" ], "title": "Summary of Dividend Applications & Payments", "organization": "State of Alaska Department of Revenue, Permanent Fund Dividend Division", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:36:30.4784085Z", "reuse": [ { "family": "resident-dividend-per-person-payment-scale", "domains": [ "economic" ], "ids": [ "2039-3" ] } ], "identity": "dd96349b8b8a1039354006b27d04ddc9fb036aad27ff87c2b55f524f2e127258", "reviewSha256": "1c013d75cd3711bcd3c2a1262f63927c3deafa2af72bd0393455b723b2958467", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:19.095Z", "lastVerifiedAt": "2026-09-26T21:05:19.095Z", "textSha256": "de5bfe1b6999b244710ba9a3d095a946c5f6965929a75f879f177cef84ab97c4", "reviewSha256": "1c013d75cd3711bcd3c2a1262f63927c3deafa2af72bd0393455b723b2958467", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-74b46d9019c3": { "url": "https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng", "urls": [ "https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng" ], "title": "Regulation (EU) 2024/1689 — Artificial Intelligence Act", "organization": "European Parliament and Council of the European Union", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2024-07-12", "dateEvidence": "OJ L, 2024/1689, 12.7.2024", "retrievedAt": "2026-09-05T00:40:27.113Z", "reuse": [ { "family": "eu-systemic-model-statutory-evaluation", "domains": [ "governance" ], "ids": [ "2026-4" ] } ], "identity": "1cda5af5520a671f2d4e223ad7f49d056232e0a52700aa62d93bb5f1a0fdc40f", "reviewSha256": "ac84674b6d55f58b9616015c6e39b9bac5e5a5d39475076b422898ed1234ab31", "health": { "status": "unavailable", "lastCheckedAt": "2026-09-26T21:05:19.778Z", "lastVerifiedAt": "2026-09-07T22:07:45.581Z", "textSha256": "fc8a96207a2baa830facc25bf6a196d2d88b04f76d8f08ee64fa22e5853b52e5", "reviewSha256": "ac84674b6d55f58b9616015c6e39b9bac5e5a5d39475076b422898ed1234ab31", "error": "Source HTTP 202.", "etag": null, "lastModified": null, "retryAt": null } }, "ref-138b13067c46": { "url": "https://arxiv.org/html/2402.08797", "urls": [ "https://arxiv.org/html/2402.08797" ], "title": "Computing Power and the Governance of Artificial Intelligence", "organization": "Sastry et al., primary research authors via arXiv", "quality": "author-manuscript", "transport": "https", "format": "html", "publishedAt": "2024-02-13", "dateEvidence": "arXiv:2402.08797v1 [cs.CY] 13 Feb 2024", "retrievedAt": "2026-09-05T00:40:23.898Z", "reuse": [ { "family": "privacy-preserving-training-workload-accounting", "domains": [ "governance" ], "ids": [ "2026-5", "2029-2" ] } ], "identity": "477ad1fb1f97c15a36b9b5958c857b62ee180190d8e987da1c4ce4dee72a9626", "reviewSha256": "a7f8749f97e7c7e17ce932ceba9d1d9355268144fe3c531b40e5ef261bc0fd29", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:19.812Z", "lastVerifiedAt": "2026-09-05T09:50:57.461Z", "textSha256": "5ad96afe01422d228a64b3a23eee580a85cca18cfe0757e0e39e8518042c4e09", "reviewSha256": "a7f8749f97e7c7e17ce932ceba9d1d9355268144fe3c531b40e5ef261bc0fd29", "error": null, "etag": "\"CJTSwdv9uZYDEAE=\"", "lastModified": "Mon, 24 Aug 2026 19:18:55 GMT", "retryAt": null } }, "ref-2543599a4374": { "url": "https://www.govinfo.gov/content/pkg/BILLS-118s3312is/html/BILLS-118s3312is.htm", "urls": [ "https://www.govinfo.gov/content/pkg/BILLS-118s3312is/html/BILLS-118s3312is.htm" ], "title": "S. 3312 Introduced in Senate — Artificial Intelligence Research, Innovation, and Accountability Act of 2023", "organization": "United States Senate; US Government Publishing Office", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2023-11-15", "dateEvidence": "November 15, 2023", "retrievedAt": "2026-09-05T00:40:24.123Z", "reuse": [ { "family": "us-federal-ai-accountability-legislative-proposal", "domains": [ "governance" ], "ids": [ "2027-4" ] } ], "identity": "83c8bcd107bd934ce4a03b0177e4098c71ea5c1caf94435afb5bd3b000c11fe7", "reviewSha256": "34df3876db05965e1356aa3e0bc57a5fd9a591df86b221ee30c21feafce25956", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:20.387Z", "lastVerifiedAt": "2026-09-05T01:10:59.268Z", "textSha256": "73a6ea393ef9812668b1d81877e99b5ab863a3d056a8158de65704285fc8e79d", "reviewSha256": "34df3876db05965e1356aa3e0bc57a5fd9a591df86b221ee30c21feafce25956", "error": null, "etag": null, "lastModified": "Tue, 28 Nov 2023 03:35:08 GMT", "retryAt": null } }, "ref-ba0e38a5fade": { "url": "https://theaipi.org/poll-biden-ai-executive-order-10-30-7-2-3/", "urls": [ "https://theaipi.org/poll-biden-ai-executive-order-10-30-7-2-3/" ], "title": "New Poll: Voters Prefer Candidates Who Want to Regulate AI Over Those Who Take Hands-Off Approach", "organization": "AI Policy Institute (AI-oversight advocacy think tank and polling publisher)", "quality": "primary-reporting", "transport": "https", "format": "html", "publishedAt": "2024-01-24T03:38:45+00:00", "dateEvidence": "2024-01-24T03:38:45+00:00", "retrievedAt": "2026-09-05T09:54:56.0660077+00:00", "reuse": [ { "family": "ai-specific-voter-salience-and-candidate-preferences", "domains": [ "governance" ], "ids": [ "2028-0" ] } ], "identity": "763e8aca8db3d9c804c54e84493ab9fd3fce729027bf6f437c576cda03860f15", "reviewSha256": "7b61c410bdc4e0e5d1acc9cb781dff33920af5150c814e9bdc221e7cb92c0cb8", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:21.889Z", "lastVerifiedAt": "2026-09-26T21:05:21.889Z", "textSha256": "10344906106d69b8cce66ec97e07d2ef33991ff819b872382f66db791e4f1413", "reviewSha256": "7b61c410bdc4e0e5d1acc9cb781dff33920af5150c814e9bdc221e7cb92c0cb8", "error": null, "etag": null, "lastModified": "Sat, 26 Sep 2026 19:38:59 GMT", "retryAt": null } }, "ref-cd4d54c88f29": { "url": "https://hai.stanford.edu/ai-index/2026-ai-index-report", "urls": [ "https://hai.stanford.edu/ai-index/2026-ai-index-report" ], "title": "The 2026 AI Index Report", "organization": "Stanford Institute for Human-Centered Artificial Intelligence", "quality": "official-research", "transport": "browser", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05", "reuse": [ { "family": "industrial-model-development-concentration", "domains": [ "geopolitical" ], "ids": [ "2028-3" ] } ], "identity": "8903467a20b2e9f28d9d14442d8a6dd9d8482812873a3bb66ea2808a81c6e268", "reviewSha256": "335bbde42a36261c5bd9493c69d662a6dc503557c8960e733d28c72b30fd53db", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:23.383Z", "lastVerifiedAt": "2026-09-26T21:05:23.383Z", "textSha256": "a064d4df6e0fb391c4d714bfee262effb74460c7e84c4a6996ef7536b9dcad7b", "reviewSha256": "335bbde42a36261c5bd9493c69d662a6dc503557c8960e733d28c72b30fd53db", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-4c0cce743341": { "url": "https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration/the-bletchley-declaration-by-countries-attending-the-ai-safety-summit-1-2-november-2023", "urls": [ "https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration/the-bletchley-declaration-by-countries-attending-the-ai-safety-summit-1-2-november-2023" ], "title": "The Bletchley Declaration by Countries Attending the AI Safety Summit, 1–2 November 2023", "organization": "UK Government and participating AI Safety Summit governments", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2023-11-01T11:30:13Z", "dateEvidence": "2023-11-01T11:30:13+00:00", "retrievedAt": "2026-09-05T00:40:28.997Z", "reuse": [ { "family": "international-frontier-safety-diplomatic-cooperation", "domains": [ "geopolitical" ], "ids": [ "2028-6" ] } ], "identity": "253eea40875b5bf0c03ecaa2fcf46613f1abaaf523c24da843db66887b49a7d8", "reviewSha256": "817b93c3c691c7793d6900b1fe21a028c6873e5ddaaa7fe3007fd29fffda7eb5", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:24.523Z", "lastVerifiedAt": "2026-09-26T21:05:24.523Z", "textSha256": "25685c614d7ecafb22f5543981e0c668608bdb1f67c8749586b057b429a695aa", "reviewSha256": "817b93c3c691c7793d6900b1fe21a028c6873e5ddaaa7fe3007fd29fffda7eb5", "error": null, "etag": "W/\"1ca7191acc71e20450f8c898a5cd96d7\"", "lastModified": null, "retryAt": null } }, "ref-67e1b50f01a0": { "url": "https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2024/05/15/statement-from-nsc-spokesperson-adrienne-watson-on-the-u-s-prc-talks-on-ai-risk-and-safety-2/", "urls": [ "https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2024/05/15/statement-from-nsc-spokesperson-adrienne-watson-on-the-u-s-prc-talks-on-ai-risk-and-safety-2/" ], "title": "Statement from NSC Spokesperson Adrienne Watson on the U.S.-PRC Talks on AI Risk and Safety", "organization": "The White House, National Security Council", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2024-05-15T13:55:56Z", "dateEvidence": "2024-05-15T13:55:56+00:00", "retrievedAt": "2026-09-05T00:40:27.710Z", "reuse": [ { "family": "us-china-intergovernmental-ai-risk-dialogue", "domains": [ "geopolitical" ], "ids": [ "2029-1" ] } ], "identity": "8ddef3d194ed1fdc4d877ee035d39043b8c0a7f765ade2457c4a36bc2ef4e232", "reviewSha256": "f31627f1d03742d4e0063af34b4e9b8056a7e848f10e14c4f7489c3556f48511", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:24.775Z", "lastVerifiedAt": "2026-09-26T21:05:24.775Z", "textSha256": "3b1cba481c99728e4f470f9552c4f7a35b02444a282a1ec84792658ee313cf3c", "reviewSha256": "f31627f1d03742d4e0063af34b4e9b8056a7e848f10e14c4f7489c3556f48511", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-5f706698d30d": { "url": "https://www.coe.int/en/web/artificial-intelligence/the-framework-convention-on-artificial-intelligence", "urls": [ "https://www.coe.int/en/web/artificial-intelligence/the-framework-convention-on-artificial-intelligence" ], "title": "The Framework Convention on Artificial Intelligence", "organization": "Council of Europe", "quality": "official-agency", "transport": "browser", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:42:04.466Z", "reuse": [ { "family": "multilateral-ai-framework-convention-participation", "domains": [ "geopolitical" ], "ids": [ "2029-4" ] } ], "identity": "7dc91b19ae7948a3703d8447cb02c943ce1b50341ae05928b378e1ef187eacac", "reviewSha256": "40896d488e0180fcaf00ee41b655b5830996362dfd1f50bbcf703134ed761131", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:32.682Z", "lastVerifiedAt": "2026-09-26T21:05:32.682Z", "textSha256": "2a4d5885559eb58f95096dc16aba8876026aa10e1bcbf593a7f067ebc7d08ca7", "reviewSha256": "40896d488e0180fcaf00ee41b655b5830996362dfd1f50bbcf703134ed761131", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-c6a691a786dc": { "url": "https://www.gov.uk/government/news/sovereign-ai-invests-in-uk-startup-reinventing-ai-chips", "urls": [ "https://www.gov.uk/government/news/sovereign-ai-invests-in-uk-startup-reinventing-ai-chips" ], "title": "Sovereign AI invests in UK startup reinventing AI chips", "organization": "UK Government", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2026-08-03T09:23:01+01:00", "dateEvidence": "Published: 3 August 2026", "retrievedAt": "2026-09-05T00:40:28.993Z", "reuse": [ { "family": "sovereign-ai-equity-investment", "domains": [ "governance" ], "ids": [ "2029-5" ] } ], "identity": "66cbbb0151b1a0c673476f5fbb4a7f24617ce81354eb87df3daab9320dcb0ada", "reviewSha256": "ad1cb801d3dfc6dd107484dd7be15030ec52132fed1f80ddaec7a78c5d40de38", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:32.923Z", "lastVerifiedAt": "2026-09-26T21:05:32.923Z", "textSha256": "e7892bf3df822ad5831530175dc56e0fbee74174b53e914c0784a5b567802c5f", "reviewSha256": "ad1cb801d3dfc6dd107484dd7be15030ec52132fed1f80ddaec7a78c5d40de38", "error": null, "etag": "W/\"fc84efee9d950e1229f142026b37a2ac\"", "lastModified": null, "retryAt": null } }, "ref-4487a62bbc1c": { "url": "https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024", "urls": [ "https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024" ], "title": "Frontier AI Safety Commitments, AI Seoul Summit 2024", "organization": "UK and Republic of Korea governments", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2024-05-21T10:00:01+01:00", "dateEvidence": "2024-05-21T10:00:01+01:00", "retrievedAt": "2026-09-05T00:40:28.989Z", "reuse": [ { "family": "frontier-risk-disclosure-confidentiality-limits", "domains": [ "governance" ], "ids": [ "2030-2" ] } ], "identity": "b76b8d9be52d212664ad2fd0f9806cb4e471286c0fff2c2cc7145aa36dae1835", "reviewSha256": "ae5826d95b5e4882fa52a952cc061d8be46a610f7e03dd677fa87137f6a36a9a", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:33.132Z", "lastVerifiedAt": "2026-09-26T21:05:33.132Z", "textSha256": "aa5d781b851e3886b45ffb97b1b04a7ac796d3e7945388ea5564aba023a98dcd", "reviewSha256": "ae5826d95b5e4882fa52a952cc061d8be46a610f7e03dd677fa87137f6a36a9a", "error": null, "etag": "W/\"8dae8f8d1c7fff99e1d2879656aa1af7\"", "lastModified": null, "retryAt": null } }, "ref-880eab95ee56": { "url": "https://ec.europa.eu/newsroom/dae/redirection/document/118119", "urls": [ "https://ec.europa.eu/newsroom/dae/redirection/document/118119", "https://digital-strategy.ec.europa.eu/en/news/general-purpose-ai-code-practice-now-available" ], "title": "General-Purpose AI Code of Practice — Safety and Security Chapter", "organization": "European Commission", "quality": "official-agency", "transport": "https", "format": "pdf", "publishedAt": "2025-07-10", "dateEvidence": "10 July 2025", "retrievedAt": "2026-09-05T00:42:10.965Z", "reuse": [ { "family": "systemic-model-external-safety-evaluation", "domains": [ "governance" ], "ids": [ "2030-3", "2031-3" ] } ], "dateEvidenceUrl": "https://digital-strategy.ec.europa.eu/en/news/general-purpose-ai-code-practice-now-available", "mediaType": "/", "identity": "9c5b86862a7d83fa56b111e11f3b8ea3e1f28ecf294a327f1528ec0cdd6d40db", "reviewSha256": "b41ba67f46926e8c65398cd57e79a113f4841d1e2820e11eaf3cd22c4ce3ed0c", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:37.889Z", "lastVerifiedAt": "2026-09-26T21:05:37.889Z", "textSha256": "09f9f9a80471e6b711a44f4ebdc97d6c4fcfde5f69f3a4042330a4b2de2f51ad", "reviewSha256": "b41ba67f46926e8c65398cd57e79a113f4841d1e2820e11eaf3cd22c4ce3ed0c", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-67c2eeb42649": { "url": "https://arxiv.org/html/2312.06942v3", "urls": [ "https://arxiv.org/html/2312.06942v3" ], "title": "AI Control: Improving Safety Despite Intentional Subversion", "organization": "Redwood Research", "quality": "author-manuscript", "transport": "https", "format": "html", "publishedAt": "2024-01-05", "dateEvidence": "arXiv:2312.06942v3 [cs.LG] 05 Jan 2024", "retrievedAt": "2026-09-05T00:40:27.196Z", "reuse": [ { "family": "intentional-subversion-control-evaluations", "domains": [ "governance" ], "ids": [ "2031-4" ] } ], "identity": "e4c17215eb3c64414587d65bac19884e6cc2bdfc8997204e029a64c4cbf8ab93", "reviewSha256": "7a96ba29cbc64e3239a0504f5191c7fdc750f61ed2182b8ea67ddb4ea4277a0c", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:37.923Z", "lastVerifiedAt": "2026-09-05T09:51:14.995Z", "textSha256": "2ef412e5abcd5d9b904814811ca9fcf5d4ba26938ad22357bfecbd6ea7dbd953", "reviewSha256": "7a96ba29cbc64e3239a0504f5191c7fdc750f61ed2182b8ea67ddb4ea4277a0c", "error": null, "etag": "\"CP6s58aHupYDEAE=\"", "lastModified": "Mon, 24 Aug 2026 20:02:56 GMT", "retryAt": null } }, "ref-c0a0d9ec675d": { "url": "https://www.govinfo.gov/content/pkg/FR-2023-10-25/html/2023-23055.htm", "urls": [ "https://www.govinfo.gov/content/pkg/FR-2023-10-25/html/2023-23055.htm" ], "title": "Implementation of Additional Export Controls: Certain Advanced Computing Items; Supercomputer and Semiconductor End Use; Updates and Corrections", "organization": "US Department of Commerce, Bureau of Industry and Security", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2023-10-25", "dateEvidence": "Wednesday, October 25, 2023", "retrievedAt": "2026-09-05T00:40:29.902Z", "reuse": [ { "family": "advanced-computing-export-licensing", "domains": [ "governance" ], "ids": [ "2032-4" ] } ], "identity": "a1d0e0cc06a959109daeb7074038f846ce304c915eac656270e67fc5b4343794", "reviewSha256": "e54753dc468ffea0e57ba831b66ff2b6fcaa2c3e46821519678c1caecfadbf92", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:39.245Z", "lastVerifiedAt": "2026-09-05T01:11:08.571Z", "textSha256": "e53219dc4008ea3e4b407064d15a0c9e46d835ac1b4dfa1ddb001a7fe02414f6", "reviewSha256": "e54753dc468ffea0e57ba831b66ff2b6fcaa2c3e46821519678c1caecfadbf92", "error": null, "etag": null, "lastModified": "Mon, 22 Sep 2025 22:39:54 GMT", "retryAt": null } }, "ref-ce8d6868fef6": { "url": "https://www.nber.org/papers/w23806", "urls": [ "https://www.nber.org/papers/w23806" ], "title": "Should Robots be Taxed?", "organization": "National Bureau of Economic Research", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2017-09-11", "dateEvidence": "2017/09/11", "retrievedAt": "2026-09-05T00:40:29.740Z", "reuse": [ { "family": "transitional-robot-tax-policy-analysis", "domains": [ "governance" ], "ids": [ "2032-5" ] } ], "identity": "9f25c5ba7751fac8f95584acd43ac048ede72c1181e72375bd5b6a9cd43a3afa", "reviewSha256": "3c7ae93b7bdba0ab7357d5a7a581dc7124fe5dbbcfbbefa825dd158fd1d89f2c", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:39.856Z", "lastVerifiedAt": "2026-09-26T21:05:39.856Z", "textSha256": "829457ecaeb449042ea9cc2c6bdfa21906835f06775bc935b6d97ada42f5d9f3", "reviewSha256": "3c7ae93b7bdba0ab7357d5a7a581dc7124fe5dbbcfbbefa825dd158fd1d89f2c", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-5b0fdfb5ea56": { "url": "https://www.un.org/sites/un2.un.org/files/governing_ai_for_humanity_final_report_en.pdf", "urls": [ "https://www.un.org/sites/un2.un.org/files/governing_ai_for_humanity_final_report_en.pdf" ], "title": "Governing AI for Humanity — Final Report", "organization": "United Nations High-level Advisory Body on Artificial Intelligence", "quality": "official-agency", "transport": "https", "format": "pdf", "publishedAt": null, "dateEvidence": "September 2024", "retrievedAt": "2026-09-05T00:42:07.285Z", "reuse": [ { "family": "international-ai-capacity-fund-benefit-sharing", "domains": [ "geopolitical" ], "ids": [ "2033-2" ] } ], "pdfPages": [ 1, 16, 17, 18, 65, 66 ], "publishedPeriod": "2024-09", "identity": "4851491dd4eda43f45930e3ce3c88bab403791bbc044daa408de3aec646568a2", "reviewSha256": "47817df9f49b748881024867c749238daff49fb1707b02dbc6966a6412c475f1", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:40.410Z", "lastVerifiedAt": "2026-09-05T10:04:49.917Z", "textSha256": "e75fb3e5a5e7e4651179347965158999978ce5cac95703bba785cbf1dc51356c", "reviewSha256": "47817df9f49b748881024867c749238daff49fb1707b02dbc6966a6412c475f1", "error": null, "etag": "W/\"412883-627b537954c18\"", "lastModified": "Mon, 25 Nov 2024 04:32:15 GMT", "retryAt": null } }, "ref-1ed19ea62985": { "url": "https://eur-lex.europa.eu/eli/reg/2024/900/oj/eng", "urls": [ "https://eur-lex.europa.eu/eli/reg/2024/900/oj/eng" ], "title": "Regulation (EU) 2024/900 on the transparency and targeting of political advertising", "organization": "European Parliament and Council of the European Union", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2024-03-20", "dateEvidence": "OJ L, 2024/900, 20.3.2024", "retrievedAt": "2026-09-05T00:40:30.235Z", "reuse": [ { "family": "political-advertising-personal-data-targeting-controls", "domains": [ "governance" ], "ids": [ "2033-3" ] } ], "identity": "51ebecced8522cf1d6bfe489ef83679efbfd0a7a716061e0e0e8db03da222fc3", "reviewSha256": "64fa7343699e54759696dd88edfb2066da3a95ac8fd8465622a0da0f5770d955", "health": { "status": "unavailable", "lastCheckedAt": "2026-09-26T21:05:40.971Z", "lastVerifiedAt": "2026-09-07T22:08:20.882Z", "textSha256": "7fd18775a171d47fcb8deed98c21c63e0301de0c704839b6760fc352505da577", "reviewSha256": "64fa7343699e54759696dd88edfb2066da3a95ac8fd8465622a0da0f5770d955", "error": "Source HTTP 202.", "etag": null, "lastModified": null, "retryAt": null } }, "ref-44a63fa0e787": { "url": "https://www.cnas.org/publications/reports/secure-governable-chips", "urls": [ "https://www.cnas.org/publications/reports/secure-governable-chips" ], "title": "Secure, Governable Chips", "organization": "Center for a New American Security", "quality": "official-research", "transport": "https", "format": "html", "publishedAt": "2024-01-08T10:00:00-05:00", "dateEvidence": "2024-01-08T10:00:00-05:00", "retrievedAt": "2026-09-05T00:40:29.113Z", "reuse": [ { "family": "hardware-enforced-adaptive-compute-licensing", "domains": [ "geopolitical" ], "ids": [ "2034-3", "2039-5" ] } ], "identity": "37db0e0f00940106e2afe76112ea2a7783faddafe6256adefda97a6fee4f61c1", "reviewSha256": "b2bd9191de651fe258b4baf3270d984276eb4017071ea8faf5fe2be204c2c1f6", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:41.210Z", "lastVerifiedAt": "2026-09-25T21:05:37.267Z", "textSha256": "ef4ba0c33544c8bf70b193a224087119e2075d01d361ba00e7294a4d29fd81a9", "reviewSha256": "b2bd9191de651fe258b4baf3270d984276eb4017071ea8faf5fe2be204c2c1f6", "error": null, "etag": null, "lastModified": "Fri, 25 Sep 2026 20:30:49 GMT", "retryAt": null } }, "ref-0bcbc033eb9e": { "url": "https://www.mfat.govt.nz/en/peace-rights-and-security/disarmament/autonomous-weapons-systems", "urls": [ "https://www.mfat.govt.nz/en/peace-rights-and-security/disarmament/autonomous-weapons-systems" ], "title": "Autonomous Weapon Systems", "organization": "New Zealand Ministry of Foreign Affairs and Trade", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:40:30.132Z", "reuse": [ { "family": "autonomous-weapons-binding-rule-diplomacy", "domains": [ "governance" ], "ids": [ "2034-5" ] } ], "identity": "3d5b980552c7273648fa1f114ec03b0510ac8f1bf17fb72b01c8e485927aaa6e", "reviewSha256": "9beff25a0ec13b6fca53b1a636db6a89a231c397fb789292fdde8e3cd4f32105", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:44.067Z", "lastVerifiedAt": "2026-09-26T21:05:44.067Z", "textSha256": "8f86609cab2b57440a8f620b5c3bda2d993697792867750b46337b0a8f736578", "reviewSha256": "9beff25a0ec13b6fca53b1a636db6a89a231c397fb789292fdde8e3cd4f32105", "error": null, "etag": "\"ede4073e5dd0a7827892a74187e765f4-gzip\"", "lastModified": null, "retryAt": null } }, "ref-06c10deb3670": { "url": "https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf", "urls": [ "https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf", "https://www.anthropic.com/responsible-scaling-policy" ], "title": "Responsible Scaling Policy, version 3.4", "organization": "Anthropic", "quality": "official-company", "transport": "https", "format": "pdf", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T09:55:12.063Z", "reuse": [ { "family": "current-responsible-scaling-policy-governance", "domains": [ "governance" ], "ids": [ "2035-1", "2040-1" ] } ], "revisionIndex": { "url": "https://www.anthropic.com/responsible-scaling-policy", "version": "3.4" }, "identity": "9407b80d183c87745960ef6334a9fc3920b7a52a576a434ad4c0461bdca9bfce", "reviewSha256": "6551a986895123a3cc11e4871fd511cb2abcd90c46e9a2f818914639d0909731", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:44.767Z", "lastVerifiedAt": "2026-09-05T09:59:40.829Z", "textSha256": "d830ad40642bbc65ed30dda4339c71e3879c02e0c3c198ee5cabd42b01fc5954", "reviewSha256": "6551a986895123a3cc11e4871fd511cb2abcd90c46e9a2f818914639d0909731", "error": null, "etag": "\"7378fad9329082f79bd5cfc800f6c04e\"", "lastModified": "Wed, 08 Jul 2026 16:19:07 GMT", "retryAt": null } }, "ref-5df33d3f8ba3": { "url": "https://citizens-initiative.europa.eu/how-it-works_en", "urls": [ "https://citizens-initiative.europa.eu/how-it-works_en" ], "title": "How it works — European Citizens' Initiative", "organization": "European Commission", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": null, "dateEvidence": null, "retrievedAt": "2026-09-05T00:40:30.938Z", "reuse": [ { "family": "citizen-initiative-institutional-participation", "domains": [ "governance" ], "ids": [ "2036-5" ] } ], "identity": "744ebe633f693ff6da4cda7ae96d983415b0ff4cfb153c32b54da84e32a9fd1f", "reviewSha256": "faf9ee98ff5b76220c393a578a59dcad16150f1ded71d9defd85723d89487d4b", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:45.787Z", "lastVerifiedAt": "2026-09-26T21:05:45.787Z", "textSha256": "828c7eb5ffceae500845642051d917c821f65d2496e48cadcbc49ee6a9687dc6", "reviewSha256": "faf9ee98ff5b76220c393a578a59dcad16150f1ded71d9defd85723d89487d4b", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-9879eed41560": { "url": "https://arxiv.org/html/2402.12572v2", "urls": [ "https://arxiv.org/html/2402.12572v2" ], "title": "FairProof: Confidential and Certifiable Fairness for Neural Networks", "organization": "University of California San Diego and Stanford University research authors", "quality": "author-manuscript", "transport": "https", "format": "html", "publishedAt": "2024-07-16", "dateEvidence": "arXiv:2402.12572v2 [cs.LG] 16 Jul 2024", "retrievedAt": "2026-09-05T00:40:30.292Z", "reuse": [ { "family": "zero-knowledge-neural-model-fairness-certification", "domains": [ "governance" ], "ids": [ "2037-2" ] } ], "identity": "da95148a90d9d688867bbe830e694b9b19587984bea4c52114f5ceeae0431566", "reviewSha256": "cfae1f67dbb72ed7bc8cdc8927ded2edf1e1f5d6364cbd091abcb2943a9df789", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:45.812Z", "lastVerifiedAt": "2026-09-05T09:51:21.826Z", "textSha256": "c04bc9857ceecf3cbcec5a74a2343ed199df09fe36982bb958944062233414b8", "reviewSha256": "cfae1f67dbb72ed7bc8cdc8927ded2edf1e1f5d6364cbd091abcb2943a9df789", "error": null, "etag": "\"CIS9ma78uZYDEAE=\"", "lastModified": "Mon, 24 Aug 2026 19:12:51 GMT", "retryAt": null } }, "ref-1598ecd76131": { "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11302104/", "urls": [ "https://pmc.ncbi.nlm.nih.gov/articles/PMC11302104/" ], "title": "Lie detection algorithms disrupt the social dynamics of accusation behavior", "organization": "iScience; von Schenk, Klockmann, Bonnefon, Rahwan and Köbis", "quality": "peer-reviewed", "transport": "https", "format": "html", "publishedAt": "2024-06-27", "dateEvidence": "2024 Jun 27", "retrievedAt": "2026-09-05T00:40:30.840Z", "reuse": [ { "family": "experimental-text-deception-classification", "domains": [ "governance" ], "ids": [ "2037-3" ] } ], "identity": "041773d22aedf0846cdbf11c94da2c2210c5f2a3d7b01ee9294dcc00eb8f496a", "reviewSha256": "de23de5d95b92d662bb7d7d3b66a647e1bf29e037a90844f2d4222ee6c607d17", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:46.426Z", "lastVerifiedAt": "2026-09-26T21:05:46.426Z", "textSha256": "7830383e5e2f0222a0b68e6204eb7e34f35ee946e722bdf41c4963fba7304c7e", "reviewSha256": "de23de5d95b92d662bb7d7d3b66a647e1bf29e037a90844f2d4222ee6c607d17", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-f434c5aa9b84": { "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC5034327/", "urls": [ "https://pmc.ncbi.nlm.nih.gov/articles/PMC5034327/" ], "title": "A physical zero-knowledge object-comparison system for nuclear warhead verification", "organization": "Nature Communications; Philippe, Goldston, Glaser and d'Errico", "quality": "peer-reviewed", "transport": "https", "format": "html", "publishedAt": "2016-09-20", "dateEvidence": "2016 Sep 20", "retrievedAt": "2026-09-05T00:40:30.804Z", "reuse": [ { "family": "physical-zero-knowledge-arms-control-verification", "domains": [ "geopolitical" ], "ids": [ "2037-5" ] } ], "identity": "2b5963dcb10c3e5393391ecc499ea9e52c16318bbd90979dae6203913dc4bd39", "reviewSha256": "7e5c98bc833724ec66f9e12863f9843014d7a38d9f480cb3a15fc360bf02eb7d", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:46.983Z", "lastVerifiedAt": "2026-09-26T21:05:46.983Z", "textSha256": "6793d22e89f6f105c6abf084c75ac0e9397d6eca34a4162d083b67b476e57cd6", "reviewSha256": "7e5c98bc833724ec66f9e12863f9843014d7a38d9f480cb3a15fc360bf02eb7d", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-3a3568b891c8": { "url": "https://www.gov.uk/government/news/government-built-humphrey-ai-tool-reviews-responses-to-consultation-for-first-time-in-bid-to-save-millions", "urls": [ "https://www.gov.uk/government/news/government-built-humphrey-ai-tool-reviews-responses-to-consultation-for-first-time-in-bid-to-save-millions" ], "title": "Government-built “Humphrey” AI tool reviews responses to consultation for first time, in bid to save millions", "organization": "UK Government", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2025-05-14T08:00:00+01:00", "dateEvidence": "2025-05-14T08:00:00+01:00", "retrievedAt": "2026-09-05T00:40:32.720Z", "reuse": [ { "family": "human-reviewed-public-administration-ai-trial", "domains": [ "governance" ], "ids": [ "2038-3" ] } ], "identity": "0116639a0cc9b5ed707dc49371751eb9abfc9bd2f1ef6354df54d500e78dde27", "reviewSha256": "645d74016466919e4842b325964c27d8b348a8dbd5e2df5811cb37af4ad7d26c", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:47.218Z", "lastVerifiedAt": "2026-09-26T21:05:47.218Z", "textSha256": "e3ddd27ee88c09ffe847233eef730d825359c4d7150b6378cd5c7eeded38d81f", "reviewSha256": "645d74016466919e4842b325964c27d8b348a8dbd5e2df5811cb37af4ad7d26c", "error": null, "etag": "W/\"ea306b65319264e911b0d438782367c7\"", "lastModified": null, "retryAt": null } }, "ref-5d67ca6a433b": { "url": "https://arxiv.org/html/2403.10462", "urls": [ "https://arxiv.org/html/2403.10462" ], "title": "Safety Cases: How to Justify the Safety of Advanced AI Systems", "organization": "Clymer, Gabrieli, Krueger and Larsen, primary research authors via arXiv", "quality": "author-manuscript", "transport": "https", "format": "html", "publishedAt": "2024-03-18", "dateEvidence": "arXiv:2403.10462v2 [cs.CY] 18 Mar 2024", "retrievedAt": "2026-09-05T00:40:31.033Z", "reuse": [ { "family": "safety-case-negative-evidence-limitations", "domains": [ "governance" ], "ids": [ "2038-5" ] } ], "identity": "71d2a2d476660b828725ca880573498f60726004151a7cef0760cccd33cf77de", "reviewSha256": "060fb8e3e4383bb260b08815cb3c0775c7dd4faeab54157d318f45f03a742f15", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:47.227Z", "lastVerifiedAt": "2026-09-05T09:51:24.473Z", "textSha256": "5efd265782a8e637f3da1144cf6a49b7d0f6138fe7ed43ddb78945e0245cebf0", "reviewSha256": "060fb8e3e4383bb260b08815cb3c0775c7dd4faeab54157d318f45f03a742f15", "error": null, "etag": "\"CLHZl/KTupYDEAE=\"", "lastModified": "Mon, 24 Aug 2026 20:58:08 GMT", "retryAt": null } }, "ref-4d35aec7fa96": { "url": "https://www.judiciary.uk/guidance-and-resources/artificial-intelligence-ai-judicial-guidance-october-2025/", "urls": [ "https://www.judiciary.uk/guidance-and-resources/artificial-intelligence-ai-judicial-guidance-october-2025/" ], "title": "Artificial Intelligence (AI) – Judicial Guidance (October 2025)", "organization": "Courts and Tribunals Judiciary of England and Wales", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2025-10-31T15:41:37Z", "dateEvidence": "2025-10-31T15:41:37+00:00", "retrievedAt": "2026-09-05T00:40:31.990Z", "reuse": [ { "family": "judicial-ai-assistance-governance", "domains": [ "governance" ], "ids": [ "2039-0" ] } ], "identity": "d15e2042ecf57f6b74b584218d3d582c073cad87dcb45a0ea5fab21fc7421c1a", "reviewSha256": "cc2540178274595858146f1cd5e1458719e04ece845ef1aea12175154adf8121", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:48.153Z", "lastVerifiedAt": "2026-09-26T21:05:48.153Z", "textSha256": "2ce6843bbd15f0ab82f09cf76ad74cc954d8cc4f8ccd0cc73f407c4dfb971f06", "reviewSha256": "cc2540178274595858146f1cd5e1458719e04ece845ef1aea12175154adf8121", "error": null, "etag": null, "lastModified": null, "retryAt": null } }, "ref-e636666c3ae8": { "url": "https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng", "urls": [ "https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng" ], "title": "Regulation (EU) 2016/679 — General Data Protection Regulation", "organization": "European Parliament and Council of the European Union", "quality": "official-agency", "transport": "https", "format": "html", "publishedAt": "2016-05-04", "dateEvidence": "OJ L 119, 4.5.2016, pp. 1–88", "retrievedAt": "2026-09-05T00:40:32.131Z", "reuse": [ { "family": "solely-automated-significant-decision-safeguards", "domains": [ "governance" ], "ids": [ "2039-2" ] } ], "identity": "3bde3b9ea5241cfafb9ec9c381b7bda9f5e9b53b3fed95093defb5dbed141b8f", "reviewSha256": "2358a2166ec3c795ad6bb41d0a2f2e954bdc05b52904c72a1518ec02b6a313ec", "health": { "status": "unavailable", "lastCheckedAt": "2026-09-26T21:05:48.685Z", "lastVerifiedAt": "2026-09-07T22:08:30.897Z", "textSha256": "a82e5e1024171b297d32afcebc82b3bdd178249d38f2f55b1269405660d8ed34", "reviewSha256": "2358a2166ec3c795ad6bb41d0a2f2e954bdc05b52904c72a1518ec02b6a313ec", "error": "Source HTTP 202.", "etag": null, "lastModified": null, "retryAt": null } }, "ref-d74601d91b96": { "url": "https://arxiv.org/html/1611.08219", "urls": [ "https://arxiv.org/html/1611.08219" ], "title": "The Off-Switch Game", "organization": "University of California, Berkeley research authors", "quality": "author-manuscript", "transport": "https", "format": "html", "publishedAt": "2017-06-16", "dateEvidence": "arXiv:1611.08219v3 [cs.AI] 16 Jun 2017", "retrievedAt": "2026-09-05T00:40:31.134Z", "reuse": [ { "family": "agent-incentives-to-preserve-shutdown-control", "domains": [ "governance" ], "ids": [ "2040-2" ] } ], "identity": "88bd9e39f3ec3db9afbf24add64327e67e145227acc874bf18635bfdbf5b3c84", "reviewSha256": "498ac0c4ce38b7a656952616938fbf99720eeb74385ca60bf5ca08cca182724d", "health": { "status": "verified", "lastCheckedAt": "2026-09-26T21:05:48.934Z", "lastVerifiedAt": "2026-09-05T09:51:24.681Z", "textSha256": "9285b442204561833199fb7c4dda9e314dd30969e61543219bfd9e1f4e227902", "reviewSha256": "498ac0c4ce38b7a656952616938fbf99720eeb74385ca60bf5ca08cca182724d", "error": null, "etag": "\"CIywvI7/uZYDEAE=\"", "lastModified": "Mon, 24 Aug 2026 19:25:10 GMT", "retryAt": null } } }, "items": { "2026-0": [ { "id": "2026-0", "sourceId": "ref-1eba12a4a8fa", "predictionText": "Frontier agents reliably complete multi-hour computer workflows with human review, while monitored research systems can persist for days or weeks and require trajectory-level safeguards", "predictionSha256": "3613a92c64832afc17f75a91e961f77f9e690fae18f6468b255a334296ddbce1", "predictionTextSha256": "88410c0c14200538a49389293abf03e5d3412e0c5eb6f8a804cdb234f0a8cfd8", "domain": "technology", "excerpt": "Over nearly 2,000 Claude Code sessions across two weeks, Opus 4.6 consumed 2 billion input tokens and generated 140 million output tokens, a total cost just under $20,000.", "relation": "measured", "direction": "supports-prerequisite", "facet": "A multi-week coding-agent research project sustained through repeated sessions and a human-designed testing harness", "why": "The compiler experiment directly demonstrates sustained project-level agent activity. The author describes adding regression checks, improving verifiers, and restructuring work when agents became stuck, making monitoring and trajectory management concrete rather than hypothetical.", "doesNotEstablish": "Two weeks describes the overall project across nearly 2,000 sessions, not uninterrupted autonomy by one agent or human-expert task difficulty. This deliberately difficult stress test does not establish reliable completion of representative computer workflows; substantial human harness design and unresolved failures remained.", "metric": { "value": 2, "unit": "weeks of project elapsed duration", "coverage": "A 16-agent Opus 4.6 compiler experiment spanning nearly 2,000 Claude Code sessions; not continuous individual-agent runtime or a METR time horizon.", "evidence": "Over nearly 2,000 Claude Code sessions across two weeks, Opus 4.6 consumed" }, "reuseFamily": "long-running-agent-compiler-construction", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2026-1": [ { "id": "2026-1", "sourceId": "ref-b05d2a6ff489", "predictionText": "Frontier labs spend roughly half of their compute on AI R&D, making internal deployment the main safety bottleneck", "predictionSha256": "88da01630da6bfe0baa004b65ab918c025b6d963fbab73cc65a47b5da7c563d2", "predictionTextSha256": "4bf5c96561353f910b1a7403ccc489637fd447243cb488898c37193c4fb28de7", "domain": "technology", "excerpt": "Claude is a constant collaborator but using it generally involves active supervision and validation, especially in high-stakes work—versus handing off tasks requiring no verification at all.", "relation": "measured", "direction": "context", "facet": "Internal engineering and research adoption creates a substantial supervision and validation workload", "why": "Anthropic studied its own engineers and researchers through an August 2025 survey, interviews, and internal usage data. The findings give an empirical reference for internal AI deployment and the continuing need for oversight inside a frontier laboratory.", "doesNotEstablish": "This is not a compute-accounting study. Reported work usage and productivity percentages cannot be converted into the share of compute allocated to AI R&D. It neither establishes roughly half of compute nor ranks internal deployment as the main safety bottleneck across laboratories.", "metric": { "value": 132, "unit": "survey respondents", "coverage": "Anthropic engineers and researchers surveyed in August 2025, supplemented by 53 interviews and internal usage analysis; not a representative census of frontier labs.", "evidence": "in August 2025 we surveyed 132 Anthropic engineers and researchers, conducted 53 in-depth qualitative interviews" }, "reuseFamily": "internal-lab-ai-work-supervision", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2026-2": [ { "id": "2026-2", "sourceId": "ref-d485fa5001db", "predictionText": "Frontier AI labs and infrastructure firms accumulate trillion-dollar-scale valuations and exceptional political leverage", "predictionSha256": "8e28475834c8fe7ec02b478bfcd3f83a4f40065aeb697a7307cafee6718bf9c8", "predictionTextSha256": "e9afa7f249a8d4d6c6c9c2b7ca160962ef6388184d00d40c49fe8087b4f6f746", "domain": "economic", "excerpt": "Pre-money valuation USD 730 billion", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Contractual valuation and concentrated capital backing of a frontier AI laboratory", "why": "SoftBank's investment agreement identifies a USD 730 billion pre-money valuation for OpenAI. This is a specific transaction-based valuation baseline approaching a trillion dollars, rather than an inferred valuation from revenue or model capability.", "doesNotEstablish": "USD 730 billion is not USD 1 trillion. The announcement concerns an investment agreement with closing conditions, not completion of every planned tranche. It does not measure political leverage or establish equivalent valuations across all frontier labs and infrastructure firms.", "metric": { "value": 730000000000, "unit": "USD pre-money valuation", "coverage": "OpenAI valuation specified in SoftBank's February 2026 follow-on investment agreement", "evidence": "Pre-money valuation USD 730 billion" }, "reuseFamily": "frontier-lab-contractual-equity-valuation", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2026-3": [ { "id": "2026-3", "sourceId": "ref-5ad436675449", "predictionText": "Humanoid robots move onto live factory lines in the thousands, but remain far short of general physical labor", "predictionSha256": "b9b0cb0c8c65dcc635387f0ba6632fdd5ee8ac51f239572ba72f38974df170f2", "predictionTextSha256": "e5c5e7554509007db8ea0786e1c5e2fe03bbaa45eb27e3ef1a4f0ec3a617d1a2", "domain": "technology", "excerpt": "Within 10 months, we launched full deployment on an active assembly line at the plant, running every single working day.", "relation": "deployment", "direction": "supports-prerequisite", "facet": "Manufacturer-reported humanoid sheet-metal loading on an active automotive assembly line", "why": "Figure reports an 11-month BMW Spartanburg deployment, including daily operation, more than 1,250 runtime hours and sheet-metal loading into welding fixtures. This supplies a concrete industrial deployment reference rather than a laboratory-only demonstration.", "doesNotEstablish": "This is Figure's account, not an independent audit. It supplies neither a thousands-of-robots installed count nor evidence of general physical labor capability. The stated placement-accuracy and intervention figures are targets, not demonstrated aggregate outcomes.", "metric": { "value": 1250, "unit": "runtime hours", "coverage": "Figure-reported Figure 02 operations at BMW Spartanburg; not robot count or independently audited uptime", "operator": ">", "evidence": "Across 1,250+ operational hours, Figure 02 recorded minimal hardware failures while generating critical data" }, "reuseFamily": "humanoid-automotive-sheet-metal-loading", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2026-4": [ { "id": "2026-4", "sourceId": "ref-74b46d9019c3", "predictionText": "Frontier-model release review becomes standard in the US and EU for cyber, bio and autonomy thresholds", "predictionSha256": "3e8463b40813332db90b246ff80b29f05681ea9c24ac85b035f072c3cf97428c", "predictionTextSha256": "5a21d67e37bc3083992d0c061cc3ec636bf0d39db0984004e799bb1d8292b90a", "domain": "governance", "excerpt": "perform model evaluation in accordance with standardised protocols and tools reflecting the state of the art, including conducting and documenting adversarial testing of the model", "relation": "policy", "direction": "supports-prerequisite", "facet": "EU statutory evaluation and adversarial-testing obligations for systemic-risk general-purpose models", "why": "Article 55 of the enacted 2024 text establishes a concrete EU evaluation, risk-mitigation, incident-reporting and cybersecurity baseline for systemic-risk models. This is a legal precursor to systematic frontier release assurance.", "doesNotEstablish": "It does not establish a common US–EU release-approval process or standard cyber, bio and autonomy thresholds. The original instrument was reviewed, not every subsequent amendment or 2026 compliance outcome.", "metric": null, "reuseFamily": "eu-systemic-model-statutory-evaluation", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2026-5": [ { "id": "2026-5", "sourceId": "ref-138b13067c46", "predictionText": "Compute tracking and inference-only verification move from niche research into government preparedness programs", "predictionSha256": "e009de8109b6f511149abcab4262ceb01660758df289d2f7a2c2b91da4dbefe8", "predictionTextSha256": "ca2816c01b2f1418d4a698e1606dc49344b0d5d108290001066dd30b2103284a", "domain": "governance", "excerpt": "Such a process would allow regulators to more accurately monitor and verify the usage of a data center’s compute resources.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Privacy-preserving workload accounting linked to identified data centres", "why": "The workload-monitoring section describes cryptographic logging and proof-of-learning linked to the data centre performing a training run. It directly addresses the technical visibility needed for government compute accounting.", "doesNotEstablish": "This is research, not evidence that a government preparedness programme deployed it. It does not prove comprehensive inference-only enforcement, resistance to adversarial evasion, or that authors' institutions endorse every proposal.", "metric": null, "reuseFamily": "privacy-preserving-training-workload-accounting", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2026-6": [ { "id": "2026-6", "sourceId": "ref-05665ed86ec7", "predictionText": "Signal: my call — genuine human-level AGI ships by end of 2026", "predictionSha256": "eabb4edd88333b3d1026fd3e12166821998866fcff83a65a6af1517ea35dc7a5", "predictionTextSha256": "eb9f7d16bc3cfee727b4cabf703994031e1c928b4e94929a2767ed339bb78df8", "domain": "technology", "excerpt": "High accuracy on HLE would demonstrate expert-level performance on closed-ended, verifiable questions and cutting-edge scientific knowledge, but it would not alone suggest autonomous research capabilities or \"artificial general intelligence.\"", "relation": "constraint", "direction": "context", "facet": "An explicit boundary between measured academic capability and evidence sufficient to establish AGI", "why": "HLE supplies a concrete, broad expert-question benchmark while explicitly rejecting the inference that success on it would establish AGI. This is a useful evidentiary constraint when assessing whether a claimed 2026 release meets a genuine human-level criterion.", "doesNotEstablish": "The benchmark does not define or certify AGI, demonstrate general autonomous research, or establish a shipping date. Its difficult expert questions are not themselves a necessary definition of average human-level intelligence, and this reference does not rule out the forecast.", "metric": { "value": 2500, "unit": "benchmark questions", "coverage": "The finalized HLE set spans over 100 subjects but remains a closed-ended academic evaluation, not an AGI certification.", "evidence": "The dataset consists of 2,500 challenging questions across over a hundred subjects." }, "reuseFamily": "cross-domain-closed-ended-expert-evaluation", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2026-7": [ { "id": "2026-7", "sourceId": "ref-da3f53e4c6fc", "predictionText": "Peer-reviewed intracortical BCI home use surpasses 3,800 hours for speech and cursor control in one participant; Neuralink PRIME-family trials remain safety-first", "predictionSha256": "3ec9c22a2747f5562fa237630a426a4dcb187b002fc6985dbd2089018b1e95e6", "predictionTextSha256": "9d48d752d09478fea0870a104cf2834296c8adeda0e8a69f7d4aef8831ad5be2", "domain": "technology", "excerpt": "The PRIME Study is a first-in-human early feasibility study to evaluate the initial clinical safety and device functionality of the Neuralink N1 Implant and R1 Robot", "relation": "trial", "direction": "supports-prerequisite", "facet": "PRIME's registered early-feasibility purpose and adverse-event-centered primary outcomes", "why": "The registered protocol explicitly prioritizes initial clinical safety and device functionality. Its primary outcomes are device-related and procedure-related adverse events at 12 months, grounding the safety-first clause in a specific study record.", "doesNotEstablish": "Registration is not regulatory approval, completed safety evidence, or peer review. This record does not establish 3,800 home-use hours, speech-and-cursor performance in one participant, or the objectives of every PRIME-family study.", "metric": { "value": 12, "unit": "months post-implant", "coverage": "Registered primary adverse-event outcome timeframe; not completed follow-up", "evidence": "{\"measure\":\"Device-Related Adverse Events (AE)\",\"timeFrame\":\"12 months post-implant\"}" }, "reuseFamily": "intracortical-bci-early-feasibility-safety", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2026-8": [ { "id": "2026-8", "sourceId": "ref-3b910b5ced30", "predictionText": "Orbital compute remains demonstrator-scale through 2026: launched GPUs run named AI workloads, but no operator discloses utility-scale power and cooling", "predictionSha256": "8687e3f78ef3d0567b4ea83ee325d3a66ca46e1875b72c65761a1c7f5757b580", "predictionTextSha256": "b6c3dbf4eafb3da3f3a8f6758b37c7afa992aafb6c0b01fec81646d241f87d46", "domain": "technology", "excerpt": "the first spacecraft to train an LLM - the nano GPT model from Andrej Karpathy.", "relation": "deployment", "direction": "supports-prerequisite", "facet": "Operator-reported nano GPT training on a launched H100-equipped satellite", "why": "Starcloud reports that Starcloud-1 launched in November 2025 carrying an NVIDIA H100 and subsequently trained the named nano GPT model. This is a specific claimed orbital workload rather than an unlaunched architectural proposal.", "doesNotEstablish": "The operator page does not independently verify workload execution, electrical power, cooling performance, or continuous service duration. It cannot establish that no other operator has disclosed utility-scale infrastructure. The page inconsistently names its inference model Gemini and Gemma; the relation rests on its separately named nano GPT training claim.", "metric": null, "reuseFamily": "orbital-gpu-named-workload-demonstrator", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2027-0": [ { "id": "2027-0", "sourceId": "ref-a06d967a4287", "predictionText": "Millions of AI-agent copies work continuously, generating at least $10B per month in paid digital labor", "predictionSha256": "94b2e6c24edc2a4663f6ecd0cba7a545bcd123900dea54e9629741039f5e8e24", "predictionTextSha256": "1456a017d6e92d7c4a0128aa23d00d0e8cec276538310111b0396ea60a22731c", "domain": "economic", "excerpt": "Agentforce ARR reached $800 million, up 169% year-over-year, and we’ve closed 29,000 deals, up 50% quarter-over-quarter.", "relation": "deployment", "direction": "supports-prerequisite", "facet": "An identifiable commercial revenue baseline for deployed enterprise agents", "why": "Agentforce's separately reported USD 800 million ARR and 29,000 deals establish that customers are buying an agent-specific enterprise product. This supplies a concrete commercialization prerequisite for a paid digital workforce rather than substituting total AI-company revenue.", "doesNotEstablish": "ARR is an annualized recurring-revenue measure, not observed monthly recognized revenue or wages paid to autonomous workers. It does not count concurrent agent copies or continuous operation. USD 10 billion per month would imply USD 120 billion annually, vastly beyond this product baseline.", "metric": { "value": 800000000, "unit": "USD annual recurring revenue", "coverage": "Company-reported Agentforce ARR at Salesforce fiscal 2026 year-end; not total Salesforce or combined Agentforce/Data 360 ARR", "evidence": "Agentforce ARR reached $800 million, up 169% year-over-year, and we’ve closed 29,000 deals, up 50% quarter-over-quarter." }, "reuseFamily": "salesforce-commercial-agent-work-deployment", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2027-1": [ { "id": "2027-1", "sourceId": "ref-1eba12a4a8fa", "predictionText": "Most production software is written end-to-end by AI, with humans specifying goals and reviewing exceptions", "predictionSha256": "3c16c7f116ca3e59833ba9c3c4b994908b6cbc4c43c8bcc61d7609a8c16183b2", "predictionTextSha256": "b79fdc6532d65eae4bb2eb857065ed609249845965306f758181eae93dcbbe1d", "domain": "technology", "excerpt": "Over nearly 2,000 Claude Code sessions and $20,000 in API costs, the agent team produced a 100,000-line compiler that can build Linux 6.9 on x86, ARM, and RISC-V.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Agents construct a substantial functioning software system from a human specification", "why": "The experiment moves beyond individual code suggestions to a large compiler implementation with meaningful integration tests. It provides a concrete prerequisite for humans shifting toward goals, evaluation, and exception handling.", "doesNotEstablish": "One research artifact does not measure the share of production software written by AI. The compiler was not a drop-in replacement, had inefficient output and unresolved defects, depended on human-designed tests, and used existing external components in some configurations.", "metric": { "value": 100000, "unit": "lines of compiler source code", "coverage": "One Rust-based C compiler research artifact; code volume is not a measure of production readiness, autonomy percentage, or software-market share.", "evidence": "The 100,000-line compiler can build a bootable Linux 6.9 on x86, ARM, and RISC-V." }, "reuseFamily": "long-running-agent-compiler-construction", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2027-2": [ { "id": "2027-2", "sourceId": "ref-056e0ff33675", "predictionText": "Full automation of AI R&D remains incomplete, but coding agents materially accelerate model research", "predictionSha256": "766eb6a2cf9b8325e7f4a6e4ffcb9356b2f7d920d76803fdf429d303ed348d56", "predictionTextSha256": "1bcc13d2c1f72e48f94e120a93b45746ce0254e193c544bd578db08ea11c2eb5", "domain": "technology", "excerpt": "To do so, the agent implemented novel custom CUDA kernels and experimented over different parameters of the solution, achieving a solution that runs in 0.64 milliseconds.", "relation": "measured", "direction": "supports-prerequisite", "facet": "Autonomous implementation and experimentation improve a GPU kernel used in a research-engineering task", "why": "In this 2024 RE-Bench experiment, o1-preview produced a prefix-sum implementation slightly faster than the best human expert solution. It directly demonstrates useful coding and empirical optimization within an AI-research engineering subtask.", "doesNotEstablish": "The 0.64 milliseconds is kernel execution time, not agent task-completion time or a research speedup ratio. Seven bounded environments with clear scoring do not establish complete AI R&D automation, and these historical models do not describe the September 2026 frontier.", "metric": { "value": 0.64, "unit": "milliseconds of kernel execution time", "coverage": "One RE-Bench prefix-sum solution from o1-preview; starting code took 4.74 ms and the best human expert solution took 0.67 ms.", "evidence": "the agent implemented novel custom CUDA kernels and experimented over different parameters of the solution, achieving a solution that runs in 0.64 milliseconds." }, "reuseFamily": "bounded-ml-research-engineering-evaluation", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2027-3": [ { "id": "2027-3", "sourceId": "ref-2a44176650f8", "predictionText": "Human-level AI becomes operationally disruptive across every major industry, forcing simultaneous business-model and workforce redesign", "predictionSha256": "65d27fc476e48c6e6fb98edb80a6d033c6b1899722276cdc57051718dbc67937", "predictionTextSha256": "68a00ed5e0ca6aeaf6b7de15811c9656171ccd0c4b0f9d114af8b2e2cde410d5", "domain": "economic", "excerpt": "As most occupations consist of tasks that require human input, transformation of jobs is the most likely impact of GenAI.", "relation": "constraint", "direction": "challenges", "facet": "Heterogeneous occupational exposure and retained human tasks constrain universal simultaneous disruption", "why": "The task-level study finds that one quarter of workers globally are in occupations with some GenAI exposure, with substantial differences by occupation and national income. Its retained-human-input finding cautions against extrapolating digital-task demonstrations to simultaneous wholesale redesign in every industry.", "doesNotEstablish": "Exposure is neither observed deployment nor job loss, and the study does not establish human-level general intelligence. Its transformation assessment is an interpretation of task exposure, not a measured future outcome or proof that later systems cannot overcome these constraints.", "metric": { "value": 25, "unit": "percent of global employment", "coverage": "Workers in occupations with some modeled GenAI exposure in the 2025 index; not jobs displaced or labor performed by AI", "evidence": "Globally, one in four workers are in an occupation with some GenAI exposure." }, "reuseFamily": "occupation-task-exposure-and-human-complements", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2027-4": [ { "id": "2027-4", "sourceId": "ref-2543599a4374", "predictionText": "The US advances an omnibus AI transparency, compute-tracking or frontier-accountability law", "predictionSha256": "e79546b27d2e02bda00547605425bd1ea4660c24115c9640b901a81770e5be1b", "predictionTextSha256": "54db8648c83b42c2acfd15cfa403e37594f59149e278d6e3f29986c136e22eb0", "domain": "governance", "excerpt": "To provide a framework for artificial intelligence innovation and accountability, and for other purposes.", "relation": "policy", "direction": "supports-prerequisite", "facet": "Introduced federal AI transparency and accountability legislation", "why": "The introduced bill combines generative-AI transparency, high-impact-system transparency reports, critical-impact risk assessments, certification and enforcement. It is a concrete legislative predecessor rather than a general call for regulation.", "doesNotEstablish": "An introduced bill is not enacted law. This source does not show passage, a 2027 legislative advance, frontier-compute tracking, or an operative federal frontier-release regime.", "metric": null, "reuseFamily": "us-federal-ai-accountability-legislative-proposal", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2027-5": [ { "id": "2027-5", "sourceId": "ref-7a179a48aad8", "predictionText": "Datacenter power, water and grid capacity become top-tier infrastructure and political constraints", "predictionSha256": "8b356902c0f10c41e534b447989e4a561eb5344d4f95fa6ac713f9e88ea19629", "predictionTextSha256": "b571156d71a5cba42e0490966e00c3c0c575f3a287c3efc2bd4973a55443d540", "domain": "geopolitical", "excerpt": "Electricity grids are already under strain in many places: we estimate that unless these risks are addressed, around 20% of planned data centre projects could be at risk of delays.", "relation": "constraint", "direction": "supports-prerequisite", "facet": "Grid-related delay risk for planned datacenter projects", "why": "The IEA connects datacenter expansion to constrained grid connections and estimates a material share of projects could be delayed without mitigation. This identifies a concrete infrastructure constraint on compute deployment.", "doesNotEstablish": "The estimate concerns conditional delay risk, not realized cancellations. It does not separately establish water scarcity, the political ranking of these issues, or their prevalence by 2027.", "metric": { "value": 20, "unit": "percent", "coverage": "IEA estimate of planned datacenter projects potentially at risk of delay absent mitigation", "evidence": "around 20% of planned data centre projects could be at risk of delays." }, "reuseFamily": "datacenter-capital-and-electricity-buildout", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2028-0": [ { "id": "2028-0", "sourceId": "ref-ba0e38a5fade", "predictionText": "AI becomes the largest issue in a major national election", "predictionSha256": "780cf84e8eac1fbb3c754c243048d7517dd7abd018c0ba9600650cc6f1a6c735", "predictionTextSha256": "66cae7275668e868594fd322418552c3b3fa4cc5697f1bfe35243cdb075a50b8", "domain": "governance", "excerpt": "73% of voters say they believe AI will become a more salient political issue for them as it develops, while 27% foresee it becoming a less important issue for them.", "relation": "measured", "direction": "context", "facet": "Voters' AI-specific candidate preferences and expectations of growing political importance.", "why": "Unlike a generic election-issues survey, this national poll explicitly asked about AI's future political salience and candidate positions on AI regulation. The reported 73% is a concrete historical measure of voters anticipating greater political importance for AI, not a measure of AI already outranking other issues. A newer February 2026 AIPI battleground poll asks regulatory preferences rather than salience; the older, more direct question is deliberately retained as historical context.", "doesNotEstablish": "The January 2024 poll was published by an AI-regulation advocacy organization. Self-reported expectations and prompted candidate preferences do not rank AI against other election issues, demonstrate actual voting behavior, or establish a largest-issue outcome in 2028. This is not a current 2026 electorate estimate. Only the publisher's public report and methodological summary were reviewed, not its respondent-level data or full questionnaire.", "metric": { "value": 73, "unit": "percent of surveyed voters", "coverage": "Respondents expecting AI to become more politically salient as it develops in AIPI's January 2024 national poll; not the share naming AI their most important voting issue.", "evidence": "73% of voters say they believe AI will become a more salient political issue for them as it develops, while 27% foresee it becoming a less important issue for them." }, "reuseFamily": "ai-specific-voter-salience-and-candidate-preferences", "reviewedAt": "2026-09-05T09:54:56.0660077+00:00", "reviewedBy": "Agent review under user authorization" } ], "2028-1": [ { "id": "2028-1", "sourceId": "ref-92735c509f66", "predictionText": "Most white-collar professions in leading economies revolve around supervising and coordinating AI agents", "predictionSha256": "d2e009311cf6622f5da5b9b1958d38158689c0c64dc82546287b2a7f2288efb0", "predictionTextSha256": "d7cf1bf84d86743dbb2e869c6396e17dc780e49e0186dc92d865c46246d76cd3", "domain": "economic", "excerpt": "In this paper, we study the staggered introduction of a generative AI-based conversational assistant using data from 5,179 customer support agents.", "relation": "measured", "direction": "supports-prerequisite", "facet": "Real workplace integration of human workers with an AI conversational assistant", "why": "The authors study an actual staggered workplace introduction and report a 14% average increase in issues resolved per hour in this working-paper version. This demonstrates an operational human-AI workflow and task-level productivity benefit that could motivate work redesign.", "doesNotEstablish": "Customer-support workers using an assistant are not necessarily supervising autonomous agents. One firm's support operation does not represent most white-collar professions or leading economies. The reported productivity gain is not a labor-displacement or managerial-role share.", "metric": { "value": 14, "unit": "percent increase in issues resolved per hour", "coverage": "Average estimated effect in the NBER working-paper study of 5,179 customer-support agents", "evidence": "Access to the tool increases productivity, as measured by issues resolved per hour, by 14% on average" }, "reuseFamily": "customer-support-human-ai-workflow-productivity", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2028-2": [ { "id": "2028-2", "sourceId": "ref-d05d86b6fe3b", "predictionText": "Frontier labs industrialize profession-by-profession training using expert interviews, environments and deployment data", "predictionSha256": "9d4497abc0c5bd83cb0cc2c208a61bb787e06f0f2d38f0366aba4ea2c1988146", "predictionTextSha256": "cd82c10703e099efa09466ed1d9a5a54209a109dd867e1588389481fe062e815", "domain": "technology", "excerpt": "The GDPval full set includes 1,320 specialized tasks (220 in the gold open-sourced set), each meticulously crafted and vetted by experienced professionals with over 14 years of experience on average from these fields.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Profession-specific expert task collection, reference materials, rubrics, and blind evaluation", "why": "GDPval operationalizes occupational knowledge into expert-authored tasks across 44 occupations, with realistic work products and professional grading. That infrastructure is a concrete prerequisite for systematic profession-by-profession model improvement.", "doesNotEstablish": "GDPval is principally an evaluation program, not proof of an industrialized training pipeline built from interviews and deployment telemetry. The initial evaluation is one-shot and omits much workplace ambiguity, interaction, and revision; one laboratory does not establish industry-wide adoption.", "metric": { "value": 1320, "unit": "specialized evaluation tasks", "coverage": "GDPval's full set across 44 occupations and nine industries; 220 tasks were released in the gold subset.", "evidence": "The GDPval full set includes 1,320 specialized tasks (220 in the gold open-sourced set)" }, "reuseFamily": "expert-curated-occupational-evaluation", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2028-3": [ { "id": "2028-3", "sourceId": "ref-cd4d54c88f29", "predictionText": "Control over frontier AI concentrates around a handful of US and Chinese companies, presidents and party leaders", "predictionSha256": "c12b5988efb4905d58bd2e6b9278e10d03ecc6460359af723ee7b8fb95274edb", "predictionTextSha256": "e6dc2b9437d3e445c716bc0208a0876ea339df46eaad65f811629be271baf894", "domain": "geopolitical", "excerpt": "Industry produced over 90% of notable frontier models in 2025", "relation": "measured", "direction": "context", "facet": "Industrial locus of notable frontier-model production", "why": "The current 2026 report supplies a 2025 industrial-production baseline rather than reusing the previous annual vintage. Its separate discussion of AI sovereignty says model production remains concentrated in the US and China. This grounds the corporate-development facet without equating industry share with concentration among particular firms.", "doesNotEstablish": "This is the report's notable frontier-model population, not all models, corporate market share or political command. No numerical change from the prior report's differently phrased population is asserted. It does not measure control by presidents or party leaders, identify an oligopoly, or establish the 2028 outcome. The landing page does not establish an exact publication date.", "metric": { "value": 90, "operator": ">", "unit": "percent of notable frontier models", "coverage": "Industry-produced share in 2025 as described by the 2026 AI Index; not within-industry concentration", "evidence": "Industry produced over 90% of notable frontier models in 2025" }, "reuseFamily": "industrial-model-development-concentration", "reviewedAt": "2026-09-05", "reviewedBy": "Agent review under user authorization" } ], "2028-4": [ { "id": "2028-4", "sourceId": "ref-7a179a48aad8", "predictionText": "Annual datacenter construction commitments exceed the US defense budget", "predictionSha256": "db5740740eb64cc4560fa6f8296dce784449e79d5f70f76ef65cbaa928ceac37", "predictionTextSha256": "ab8477a10e7cb3819a9345d24cfcbc3bd369b115f64fff766d978b955afc44bb", "domain": "economic", "excerpt": "Global investment in data centres has nearly doubled since 2022 and amounted to half a trillion dollars in 2024.", "relation": "measured", "direction": "context", "facet": "Historical annual global datacenter investment scale", "why": "The IEA's approximately $500 billion investment estimate supplies a concrete historical magnitude for the datacenter capital expansion underlying the forecast.", "doesNotEstablish": "Investment is not the same accounting category as annual construction commitments. The source does not supply a like-for-like US defense-budget comparison, exclude equipment expenditure, resolve geographic scope, or demonstrate the forecast's 2028 threshold.", "metric": { "value": 500000000000, "unit": "USD", "coverage": "IEA estimate of global datacenter investment in 2024; not construction-only commitments", "evidence": "Global investment in data centres has nearly doubled since 2022 and amounted to half a trillion dollars in 2024." }, "reuseFamily": "datacenter-capital-and-electricity-buildout", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2028-5": [ { "id": "2028-5", "sourceId": "ref-1b5c7b499756", "predictionText": "Superintelligence emerges and recursive self-improvement begins on my ungoverned 2028–2030 branch", "predictionSha256": "f281fbad7c594bf1d43543a112357a6a01fc506043f1cb4c1827d0b1feb8d8cd", "predictionTextSha256": "14ec17ba0fb37e06fbb8f1b3e0f58e23eac1d86cf5113664c766cdf96dbc6766", "domain": "technology", "excerpt": "AlphaEvolve enhanced the efficiency of Google's data centers, chip design and AI training processes — including training the large language models underlying AlphaEvolve itself.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "AI-generated optimization feeds back into the training infrastructure of its underlying model family", "why": "AlphaEvolve provides a concrete feedback-loop precursor: an AI coding system improves kernels involved in training the models on which that system depends. The reported Gemini training-time reduction is narrower and more informative than a generic assertion of recursive improvement.", "doesNotEstablish": "This is not demonstrated superintelligence or an autonomous, open-ended self-improvement cascade. Human-selected objectives and automated evaluators remain central, and the result does not establish an ungoverned branch, a takeoff rate, or the 2028–2030 timing.", "metric": { "value": 1, "unit": "percent reduction in Gemini training time", "coverage": "Reported consequence of a 23% speedup to one matrix-multiplication kernel; not a 1% general intelligence improvement or a full research-automation result.", "evidence": "it sped up this vital kernel in Gemini’s architecture by 23%, leading to a 1% reduction in Gemini's training time." }, "reuseFamily": "ai-generated-training-kernel-feedback", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2028-6": [ { "id": "2028-6", "sourceId": "ref-4c0cce743341", "predictionText": "International negotiations begin over slowing an intelligence explosion and auditing frontier compute", "predictionSha256": "776c3c3273a57b66cd7d75480d65d89f87e32c34db5d4f3280495669faff7be5", "predictionTextSha256": "607d10b8aa485911b7ceff93aab92a5d394d365674025b973300786baa0cff99", "domain": "geopolitical", "excerpt": "we resolve to intensify and sustain our cooperation, and broaden it with further countries, to identify, understand and as appropriate act", "relation": "precursor", "direction": "supports-prerequisite", "facet": "International diplomatic cooperation specifically addressing frontier-AI risks", "why": "The declaration identifies catastrophic frontier risks, calls for sustained international cooperation and lists both China and the United States among participants. It establishes a relevant diplomatic forum and shared risk agenda.", "doesNotEstablish": "It does not announce negotiations to slow an intelligence explosion, inspect compute, limit training or enact a treaty. Agreement to cooperate on safety is narrower than those forecast bargaining objectives.", "metric": null, "reuseFamily": "international-frontier-safety-diplomatic-cooperation", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2028-7": [ { "id": "2028-7", "sourceId": "ref-fea5cc03eeb8", "predictionText": "Peer review confirms a synapse-resolution whole-brain connectome for a vertebrate larva, without demonstrating functional emulation", "predictionSha256": "bccf963b3e78631cf3148721b5924365c7ea43ece46155b5ecb380aca0663610", "predictionTextSha256": "24cf770034939318dc78b8a0caefd524ba5add3291739f1f3d88d9b889519567", "domain": "technology", "excerpt": "the synaptic-resolution connectome of an entire brain of an insect, the Drosophila melanogaster larva, comprising 3016 neurons and 548,000 synapses has been mapped using computer-assisted reconstruction from electron micrographs.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Synaptic-resolution reconstruction of a complete insect larval brain", "why": "The participating MRC research institution reports the Drosophila larval reconstruction and explicitly links Winding and colleagues' Science paper, DOI 10.1126/science.add9330. Computer-assisted reconstruction from electron micrographs supplies a concrete methodological precursor for whole-brain larval connectomics.", "doesNotEstablish": "Drosophila is an invertebrate, not the vertebrate required by the forecast. Anatomical reconstruction and network analysis do not demonstrate functional whole-brain emulation. This institutional account summarizes the study rather than providing an independent replication or a new vertebrate result.", "metric": { "value": 3016, "unit": "neurons", "coverage": "Drosophila melanogaster larval whole-brain reconstruction reported by the participating institution; not a vertebrate brain", "evidence": "the Drosophila melanogaster larva, comprising 3016 neurons and 548,000 synapses" }, "reuseFamily": "larval-whole-brain-synaptic-reconstruction", "reviewedAt": "2026-09-05T00:50:51.848Z", "reviewedBy": "Agent review under user authorization" } ], "2029-0": [ { "id": "2029-0", "sourceId": "ref-52dbf968dc6c", "predictionText": "AI systems perform at least one quarter of cognitive labor in a leading economy", "predictionSha256": "71a1f98b4ecb4ae8cafd278335893c366812208cd612ebabc2f370fd5bb45d9b", "predictionTextSha256": "2d1721bca60ad333e226f7276d7ae8d4371ba408311002a315fbafba964b5848", "domain": "economic", "excerpt": "We estimated that between 0.5% and 3.5% of all work hours in the U.S. are currently assisted by generative AI.", "relation": "measured", "direction": "context", "facet": "Economy-wide work-hour intensity of actual reported generative-AI assistance", "why": "The nationally representative survey offers a time-intensity baseline rather than counting users or exposed occupations. Its August 2024 estimate places AI assistance in 0.5%–3.5% of all U.S. work hours, illustrating the measurement and adoption gap before claiming a quarter of cognitive labor.", "doesNotEstablish": "Assisted work hours are not hours performed autonomously by AI. All work hours and cognitive-labor hours have different denominators, so this range cannot be directly subtracted from 25%. The survey is self-reported and does not measure AI's share of economic value.", "metric": { "value": 0.5, "unit": "percent of all U.S. work hours", "coverage": "Estimated hours assisted by generative AI using the August 2024 Real-Time Population Survey", "high": 3.5, "evidence": "We estimated that between 0.5% and 3.5% of all work hours in the U.S. are currently assisted by generative AI." }, "reuseFamily": "survey-measured-ai-assisted-work-hour-intensity", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2029-1": [ { "id": "2029-1", "sourceId": "ref-67e1b50f01a0", "predictionText": "The US and China enter serious negotiations on frontier-compute declarations, inspections and training limits", "predictionSha256": "3f1263079d8495ce2345e9b38b1cc457be37afdc9f48202eaaf6a71998a69de4", "predictionTextSha256": "67c0b86e031fa5eb3ff54f91b801854c19afc88b35ea4a401a55c237f1b92a3f", "domain": "geopolitical", "excerpt": "In a candid and constructive discussion, the United States and PRC exchanged perspectives on their respective approaches to AI safety and risk management.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "An actual bilateral US–China governmental AI-risk dialogue", "why": "The official account documents interagency US and PRC delegations meeting in Geneva on 14 May 2024 about AI risk and safety. This is the bilateral communication channel needed before more intrusive compute negotiations could occur.", "doesNotEstablish": "The disclosed agenda contains no compute declarations, inspection protocol, training limit or negotiated bargain. Dialogue and risk-management exchanges do not establish serious negotiations on those specific controls.", "metric": null, "reuseFamily": "us-china-intergovernmental-ai-risk-dialogue", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2029-2": [ { "id": "2029-2", "sourceId": "ref-138b13067c46", "predictionText": "Managed branch: the US and China temporarily pause the largest frontier training runs while preserving inference", "predictionSha256": "a7a629588f6013eed9adecc3ea783a1a06e9e08172405a1a363a08e37721c6de", "predictionTextSha256": "20aa7c56ffd9e90c1799da9dd78d1ab3c79bbbcf93c3052d6253931139c90e72", "domain": "governance", "excerpt": "It also provides a clearer distinction between the compute resources used for training purposes and those that were not.", "relation": "feasibility", "direction": "supports-prerequisite", "facet": "Distinguishing training from other workloads for selective restrictions", "why": "The workload-accounting proposal links training-compute proofs to data centres and explicitly discusses distinguishing training resources from other use. That distinction is technically relevant to restricting training without indiscriminately disabling inference.", "doesNotEstablish": "No US–China pause, agreed training threshold or inference-preserving enforcement deployment is reported. Accounting proposals do not prove complete detection of concealed training or robust enforcement against hostile operators.", "metric": null, "reuseFamily": "privacy-preserving-training-workload-accounting", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2029-3": [ { "id": "2029-3", "sourceId": "ref-443d9796ce11", "predictionText": "Inference-only verification is deployed at multiple major datacenters", "predictionSha256": "9e21e4f01d8b6e1738d1350244c1f95e2cdf29f44c510f92d64a9335953aed43", "predictionTextSha256": "e3f6f645bc0c490d3332954f33eab8134cd15e54ea00dfcd83edc39b8e197b89", "domain": "technology", "excerpt": "The user of the confidential computing environment can check the attestation report and only proceed if it is valid and correct.", "relation": "feasibility", "direction": "supports-prerequisite", "facet": "Hardware-rooted, cryptographically signed accelerator-state attestation before admitting a workload", "why": "NVIDIA describes an implemented H100 chain of trust, signed measurements, and local or remote attestation verification. This could supply a trusted-state substrate on which a future inference-only enforcement and verification system is built.", "doesNotEstablish": "Device and firmware attestation does not prove that a workload is inference rather than training. The source explicitly supports both training and inference. It does not demonstrate an inference-only verifier, its soundness, or its deployment at multiple datacenters; the cited release was early access in 2023.", "metric": null, "reuseFamily": "accelerator-state-attestation-prerequisite", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2029-4": [ { "id": "2029-4", "sourceId": "ref-5f706698d30d", "predictionText": "A multilateral AI consortium or treaty framework gains support beyond the US and China", "predictionSha256": "336fc331c268a07526b5ba07714ac106cf4e3485b98150ebcb8d0f81578e14d0", "predictionTextSha256": "3da6bb9afe8dbeb01f0936a0160b8599501350cc0192cb4a484ded94df2b91f1", "domain": "geopolitical", "excerpt": "The Framework Convention was drafted by the 46 member states of the Council of Europe, with the participation of all observer states:", "relation": "policy", "direction": "supports-prerequisite", "facet": "AI treaty-framework development involving countries beyond the two largest AI powers", "why": "The Council of Europe's own account documents a negotiated AI framework involving its 46 member states, observer states, the EU and additional non-member countries. This is a specific multilateral institutional precedent.", "doesNotEstablish": "The convention is not a frontier-compute consortium, a training-cap treaty or a US–China bargain. Participation in drafting does not establish ratification, entry into force for every participant, or a new 2029 coalition.", "metric": null, "reuseFamily": "multilateral-ai-framework-convention-participation", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2029-5": [ { "id": "2029-5", "sourceId": "ref-c6a691a786dc", "predictionText": "Erosion of labor-tax revenue makes AI dividends, sovereign AI stakes and compute rents mainstream policy", "predictionSha256": "a0a96449516d124072752aa381237d0873d2338afcb54ed1e19042e17e4be5f7", "predictionTextSha256": "896d2162f906f3a35ed4afcf05afed48ce03b0ae8e9e24bf7e74808d83f78ea2", "domain": "governance", "excerpt": "This backing brings the number of startups receiving equity investment from Sovereign AI to 5 since it launched", "relation": "measured", "direction": "supports-prerequisite", "facet": "Actual sovereign equity investment in AI companies", "why": "The government reports an equity investment in AI-chip company OLIX and five startups receiving Sovereign AI equity backing. This is an implemented public-ownership mechanism, not merely discussion of sovereign stakes.", "doesNotEstablish": "It does not attribute investment to labour-tax erosion or establish mainstream compute-rent taxation. Public equity ownership does not create an individual AI dividend or compensation entitlement.", "metric": { "value": 5, "unit": "startups receiving equity investment", "coverage": "UK Sovereign AI portfolio count reported on this publication date; not citizen dividend recipients or total sovereign AI ownership globally.", "evidence": "This backing brings the number of startups receiving equity investment from Sovereign AI to 5 since it launched" }, "reuseFamily": "sovereign-ai-equity-investment", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2029-6": [ { "id": "2029-6", "sourceId": "ref-aab8dd5402ac", "predictionText": "AI policy shocks cause sustained market volatility and political polarization", "predictionSha256": "b7b2df01a85a7650317014199e2ba44b1761578736578f6684b383a50880f198", "predictionTextSha256": "56b05af99eee894139eb5b9620c61e4293f36db76fa8ffbde62482a9503ba3b7", "domain": "social", "excerpt": "On April 9, 2025, NVIDIA was informed by the U.S. government that a license is required for exports of its H20 products into the China market.", "relation": "measured", "direction": "supports-prerequisite", "facet": "A discrete AI-chip policy change causing a material corporate financial shock", "why": "NVIDIA reports a USD 4.5 billion inventory and purchase-obligation charge following new H20 export-license requirements. This is a concrete financial-statement and earnings impact of an AI-chip policy change, providing one channel through which policy could affect investors' valuations; it is not an observed cash-flow loss.", "doesNotEstablish": "A charge is not a measured stock-volatility series. This source does not show sustained market volatility, political polarization, or causal links between the two. The announced future-quarter revenue outlook is not treated as an observed loss.", "metric": { "value": 4500000000, "unit": "USD charge", "coverage": "H20 excess inventory and purchase obligations recognized in NVIDIA's first quarter of fiscal 2026", "evidence": "NVIDIA incurred a $4.5 billion charge in the first quarter of fiscal 2026 associated with H20 excess inventory and purchase obligations" }, "reuseFamily": "ai-chip-export-control-earnings-shock", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2030-0": [ { "id": "2030-0", "sourceId": "ref-056e0ff33675", "predictionText": "Absent a sustained slowdown, AI fully automates frontier AI R&D by 2030", "predictionSha256": "e3e05e382ea75e1214e9508e15082477aece36de05c7e573c1f8a93be661bf81", "predictionTextSha256": "b0e3ff985b6c9077344f58455b2814842c4c3bf57780fe8960b01a94a456a265", "domain": "technology", "excerpt": "At a 32-hour time budget, the average human score is almost twice that of the best AI agent.", "relation": "constraint", "direction": "challenges", "facet": "Research-engineering performance fails to scale with larger task budgets as effectively as human performance", "why": "This historical comparison identifies a specific obstacle between fast local coding and complete research automation: humans gained more from additional experimentation time. METR also distinguishes its clear, fast-feedback tasks from real research with ambiguous goals and long feedback loops.", "doesNotEstablish": "The comparison concerns November 2024 agents, not current frontier performance, and cannot establish that automation will remain incomplete in 2030. Its 32-hour number is a total computer-time allocation across attempts, not a METR task-difficulty time horizon or uninterrupted agent runtime.", "metric": { "value": 32, "unit": "hours of total computer-time budget", "coverage": "RE-Bench's seven bounded ML-research engineering environments with best-observed attempt allocations; average human scores were almost twice the best agent's.", "evidence": "At a 32-hour time budget, the average human score is almost twice that of the best AI agent." }, "reuseFamily": "bounded-ml-research-engineering-evaluation", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2030-1": [ { "id": "2030-1", "sourceId": "ref-1f65760bea2b", "predictionText": "AI 2040 default branch: top-expert or superintelligent AI follows automated coding within roughly one year", "predictionSha256": "183b5e3a05473ea9a2a3be4e37cb193531faccf2b4907194a8492210fef91cd2", "predictionTextSha256": "3e88bf59f4681877aff6f98cbf49ac6504b426e5d71848b1438da902466bcbd4", "domain": "technology", "excerpt": "An advanced version of Gemini Deep Think solved five out of the six IMO problems perfectly, earning 35 total points, and achieving gold-medal level performance.", "relation": "measured", "direction": "supports-prerequisite", "facet": "High-level non-coding mathematical reasoning with externally graded natural-language solutions", "why": "The result supplies a concrete reasoning milestone beyond software construction. IMO coordinators graded the submitted solutions, providing a specific empirical prerequisite for broader cognitive capability rather than relying on the scenario's assumed transition.", "doesNotEstablish": "IMO evaluates elite pre-university mathematics, not research-level mastery across cognitive fields. The graders did not validate the underlying model or process. This result establishes neither automated coding as a cause nor an approximately one-year transition to top-expert AI or superintelligence.", "metric": { "value": 35, "unit": "IMO points", "coverage": "Five of six IMO 2025 problems solved by a specially enhanced Gemini Deep Think system; gold-medal-standard solutions, not general expert equivalence.", "evidence": "solved five out of the six IMO problems perfectly, earning 35 total points" }, "reuseFamily": "externally-graded-olympiad-reasoning", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2030-2": [ { "id": "2030-2", "sourceId": "ref-4487a62bbc1c", "predictionText": "Managed branch: frontier R&D resumes under total research transparency and cross-border verification", "predictionSha256": "018efdacd8e709ef13097b57e17fe127c255972e8f7421184f010025d1a6b3e1", "predictionTextSha256": "141b8f9fbe4fd2a90f5cf6278ebe51be63d2bdaa6f2e8e5850f7c8b4bd15699e", "domain": "governance", "excerpt": "They should still share more detailed information which cannot be shared publicly with trusted actors, including their respective home governments or appointed body, as appropriate.", "relation": "constraint", "direction": "challenges", "facet": "Security and commercial limits on comprehensive research disclosure", "why": "Commitment VII distinguishes public transparency from sensitive information shared with trusted actors. Its exceptions show that an existing international frontier-safety mechanism deliberately stops short of total research transparency.", "doesNotEstablish": "The commitments are voluntary and do not create cross-border inspections, a training pause or its resumption. Confidential disclosure to a home government is not verification by another country, and the exceptions do not prove future stronger arrangements impossible.", "metric": null, "reuseFamily": "frontier-risk-disclosure-confidentiality-limits", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2030-3": [ { "id": "2030-3", "sourceId": "ref-880eab95ee56", "predictionText": "Managed branch: algorithms are broadly auditable while frontier weights remain controlled against misuse", "predictionSha256": "e13d471c2c554636b731c1023490bafad327fb7379d499cb9ebc8edd92c3ad55", "predictionTextSha256": "60b1d3c85d8f01d07322006014e4234570144c10d02bec24cb18aff082aab009", "domain": "governance", "excerpt": "Signatories will provide an adequate number of independent external evaluators with adequate free access to:", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Independent evaluation access without requiring public release of model weights", "why": "Page 14 specifies independent evaluator access and permits API, on-premise and provider-hardware access routes. These are concrete mechanisms for outside scrutiny while retaining custody of non-public model parameters.", "doesNotEstablish": "Evaluator access does not make algorithms generally interpretable or broadly auditable. The voluntary code does not establish universal weight controls, complete evaluator access, a managed-branch settlement or public release of every audit.", "metric": null, "reuseFamily": "systemic-model-external-safety-evaluation", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2030-4": [ { "id": "2030-4", "sourceId": "ref-16ebffb99c27", "predictionText": "The first drugs substantially designed by AI gain major-regulator approval", "predictionSha256": "bdcb3fcade430e3f79b22c1ce5bd84a0635fc98b585d48e662af6ea131cea9bb", "predictionTextSha256": "84182ab838404e49349c2bcbb19da1dfe06c9c6d4da8f4fedfde46a56b8c1918", "domain": "technology", "excerpt": "Participants were randomly assigned to receive either placebo, 30 mg Rentosertib once daily (QD), 30 mg twice daily (BID), or 60 mg QD for 12 weeks.", "relation": "trial", "direction": "supports-prerequisite", "facet": "Developer-reported randomized human testing of an AI-designed small molecule", "why": "Insilico describes the same Nature Medicine phase IIa GENESIS-IPF study: 71 patients across 22 Chinese sites, randomized among placebo and three rentosertib dosing groups for 12 weeks. The developer attributes both target identification and molecular design to AI. This is a specific sponsor-reported clinical-development prerequisite rather than an approval claim.", "doesNotEstablish": "This is the developer's account, not independent verification; its favorable safety and efficacy characterizations must not be treated as established clinical benefit. A small, 12-week phase IIa trial is not major-regulator approval, successful phase III testing, or evidence of durable efficacy. The announcement itself acknowledges limited group sizes and the need for larger studies. It does not establish which AI-designed drug will be approved first or when.", "metric": { "value": 71, "unit": "patients enrolled", "coverage": "Company-reported enrollment in the four-arm, 12-week GENESIS-IPF phase IIa study across 22 sites in China", "evidence": "a double-blind, placebo-controlled trial that enrolled 71 patients with IPF across 22 sites in China." }, "reuseFamily": "ai-designed-small-molecule-clinical-development", "reviewedAt": "2026-09-05T00:50:51.848Z", "reviewedBy": "Agent review under user authorization" } ], "2030-5": [ { "id": "2030-5", "sourceId": "ref-7a179a48aad8", "predictionText": "Physical production, energy and robotics—not ideas—become the main bottlenecks to AI-driven growth", "predictionSha256": "545d5ec1258351678405750bb8cd3ddcbf13205c522af6a4c4189669ae54b2bb", "predictionTextSha256": "7fce36ac55317449539da7894d328a34b46db1dd000002ddf2b83947ac2e0db6", "domain": "economic", "excerpt": "Building new transmission lines can take four to eight years in advanced economies and wait times for critical grid components such as transformers and cables have doubled in the past three years.", "relation": "constraint", "direction": "supports-prerequisite", "facet": "Physical grid-construction and equipment lead times constrain compute expansion", "why": "Transmission construction and transformer or cable delivery require long physical production cycles. These constraints can persist even when AI algorithms improve rapidly, supplying a specific mechanism through which energy infrastructure limits deployment.", "doesNotEstablish": "The source does not rank physical inputs against ideas as economy-wide growth bottlenecks, demonstrate robotics scarcity, or establish that physical constraints dominate AI-driven growth in 2030.", "metric": null, "reuseFamily": "datacenter-capital-and-electricity-buildout", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2031-0": [ { "id": "2031-0", "sourceId": "ref-056e0ff33675", "predictionText": "On the unpaused branch, fully automated AI R&D delivers roughly a 10x research speedup", "predictionSha256": "02dada827f3c40b87871578318c66f1a503116f5314a25545bb0b54e81a14011", "predictionTextSha256": "51bbd0773d7a4c17b843ae25b5e1d4909c99a0fef00a6c9979bed2eb6b7c6b69", "domain": "technology", "excerpt": "They generate and test implementations more than ten times faster than humans, allowing them to sometimes find very strong solutions and make consistent progress when given more attempts.", "relation": "measured", "direction": "context", "facet": "More-than-tenfold implementation and testing throughput within bounded ML-research engineering tasks", "why": "The experiment measures a local mechanism that could contribute to faster research: agents can generate and test many implementations quickly. It provides a quantitative reference while separating implementation throughput from useful research output.", "doesNotEstablish": "This is not a tenfold increase in quality-adjusted discoveries or end-to-end frontier research. The same study finds weaker agent performance at larger budgets. It does not establish complete automation, an unpaused policy branch, or an aggregate speedup in 2031.", "metric": { "value": 10, "unit": "times human implementation-generation and testing throughput", "coverage": "Historical 2024 RE-Bench observations in seven research-engineering environments; not a full research-cycle speedup.", "operator": ">", "evidence": "They generate and test implementations more than ten times faster than humans" }, "reuseFamily": "bounded-ml-research-engineering-evaluation", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2031-1": [ { "id": "2031-1", "sourceId": "ref-9c5b9fb39a11", "predictionText": "AI performs about one third of cognitive labor and robots about one tenth of physical labor in a leading economy", "predictionSha256": "0652fb973669821703efbeb4d3e0a98e385989de2f16b7b2d766c1920fa5792d", "predictionTextSha256": "edcfa575cb2ce3a2af93e8a6931422dde10b8754c33353782d46f8ca8686f227", "domain": "economic", "excerpt": "The United States ranks 8th worldwide with 307 units per 10,000 employees.", "relation": "measured", "direction": "context", "facet": "Installed industrial-robot density as a physical-automation baseline in a leading economy", "why": "IFR's World Robotics 2025 statistics, reported in its April 8, 2026 release, identify 307 operational industrial robots per 10,000 manufacturing employees in the United States in 2024. This is a concrete installed-capital baseline for physical automation rather than a projection of humanoid deployment.", "doesNotEstablish": "Robot counts per employee cannot be converted into physical-labor shares without utilization, task-time and productivity data. Manufacturing excludes much physical work in services, construction and agriculture. The source does not measure cognitive labor performed by AI. The published definition is compatible with the earlier baseline, but an unrevised US employment-denominator time series is not established; no growth-rate claim is inferred.", "metric": { "value": 307, "unit": "operational industrial robots per 10,000 manufacturing employees", "coverage": "United States, 2024; IFR World Robotics 2025 statistics reported April 8, 2026", "evidence": "The United States ranks 8th worldwide with 307 units per 10,000 employees." }, "reuseFamily": "manufacturing-industrial-robot-installed-density", "reviewedAt": "2026-09-07", "reviewedBy": "Agent review under user authorization" } ], "2031-2": [ { "id": "2031-2", "sourceId": "ref-a0af5ab6537d", "predictionText": "Combined revenue of the largest AI companies rivals the annual revenue of a major national government", "predictionSha256": "fd46bf957ede853d3c88627870994b7b120b38800c8bc653888dd540e073b8c3", "predictionTextSha256": "1f84c7e47279cac83fe6b039020b13ee17230d5fc6e568c28f613c04fd37af4d", "domain": "economic", "excerpt": "Revenue was $331.8 billion and increased 18% (up 16% in constant currency)", "relation": "measured", "direction": "context", "facet": "Reported annual business scale of a major AI platform and infrastructure company", "why": "Microsoft's latest completed-fiscal-year results report USD 331.8 billion of revenue for the year ended June 30, 2026. This updates the actual company-scale baseline for a future aggregate-company versus government-receipts comparison, without substituting quarterly sales, market capitalization or investment commitments for annual revenue.", "doesNotEstablish": "This is total Microsoft revenue, not AI-only revenue. It does not aggregate the largest AI companies, specify a national-government comparator, or establish that the forecast threshold has been reached. Company sales and government receipts have different economic meanings.", "metric": { "value": 331800000000, "unit": "USD annual company revenue", "coverage": "Microsoft fiscal year ended June 30, 2026; rounded headline in the Fiscal Year 2026 Results section, not fourth-quarter revenue", "evidence": "Revenue was $331.8 billion and increased 18% (up 16% in constant currency)" }, "reuseFamily": "ai-platform-company-reported-annual-revenue", "reviewedAt": "2026-09-05T01:02:29.0459297Z", "reviewedBy": "Agent review under user authorization" } ], "2031-3": [ { "id": "2031-3", "sourceId": "ref-880eab95ee56", "predictionText": "Public, externally reviewed safety cases become a requirement for the highest-risk frontier deployments", "predictionSha256": "349bbb15522677bccbb8a74dc5daa37a52045d54c4352321e1d35bf2f66eed57", "predictionTextSha256": "2736423e28beb858ec79f50fb76aa5f596575b7ed2b78fff07e336ad2e5803b9", "domain": "governance", "excerpt": "Signatories commit to reporting to the AI Office information about their model and their systemic risk assessment and mitigation processes and measures", "relation": "policy", "direction": "supports-prerequisite", "facet": "Pre-market model assurance reports incorporating external evaluation and security-review material", "why": "Commitment 7 requires signatories to create a Safety and Security Model Report before market placement. Pages 20–21 address independent evaluation information, external security reports and explanations when external evaluation was not used.", "doesNotEstablish": "A report to the AI Office is not necessarily a fully public safety case. The code is a voluntary compliance route, includes external-evaluation conditions and confidentiality limits, and does not establish the forecast universal public-review requirement.", "metric": null, "reuseFamily": "systemic-model-external-safety-evaluation", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2031-4": [ { "id": "2031-4", "sourceId": "ref-67c2eeb42649", "predictionText": "Repeated frontier-agent circumvention, sandbox-escape or sabotage incidents make trajectory-level monitoring and externally reviewed control cases mandatory for the highest-risk deployments", "predictionSha256": "713efe97dfbadbe77e2489cec0d0c969cf63c2a921f6fabb696295222f84d78d", "predictionTextSha256": "01378ae5d871093085a295e88658b7e13a8e53c2b506dec0199a515d9fc93f46", "domain": "governance", "excerpt": "We investigate protocols that aim to never submit solutions containing backdoors, which we operationalize here as logical errors that are not caught by test cases.", "relation": "trial", "direction": "supports-prerequisite", "facet": "Adversarial evaluation of monitoring and editing protocols against intentional sabotage", "why": "The study tests trusted editing and untrusted monitoring against deliberately subversive coding strategies. It provides a concrete experimental foundation for control evaluations rather than treating ordinary benchmark accuracy as evidence of control.", "doesNotEstablish": "These are constructed coding experiments, not repeated real-world sandbox escapes or long-horizon frontier-agent incidents. The source establishes neither mandatory trajectory monitoring nor legally required external control cases.", "metric": null, "reuseFamily": "intentional-subversion-control-evaluations", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2031-5": [ { "id": "2031-5", "sourceId": "ref-58a0c016acd5", "predictionText": "Continual-learning architectures become a major regulatory flashpoint because deployed capabilities can change", "predictionSha256": "35c33849a11de7b1cd1f7686e3177c609e4df5b04d8f4fd2b152b44499d591b9", "predictionTextSha256": "519d33f3f30d9b58a2be761ffbc3e6cbaaeafdf6c18ef5b4a6d58ebf2479a9b7", "domain": "technology", "excerpt": "Instead of compressing information into a static state, this architecture actively learns and updates its own parameters as data streams in.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Test-time memory learning changes model parameters while processing incoming information", "why": "Titans and MIRAS provide a concrete architectural route to adaptation without dedicated offline retraining. That mechanism is relevant to whether a predeployment evaluation continues to characterize a system after subsequent learning.", "doesNotEstablish": "Test-time memory adaptation is not evidence of unbounded general capability growth, persistent changes to every component, or widespread deployment. This research summary does not establish regulatory controversy, legal treatment, or a major flashpoint by 2031.", "metric": null, "reuseFamily": "test-time-parameterized-memory-learning", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2032-0": [ { "id": "2032-0", "sourceId": "ref-2be84f63d7cb", "predictionText": "AI systems produce more cognitive labor than humans in at least one leading economy", "predictionSha256": "127d5386ca818d2c6e7957cb4615af07699b5acce34d347bca3c97bb8ffb9a75", "predictionTextSha256": "274a0984806c40fee1a395ec8129300bea348e4eb61a32214b54af759cb6fa75", "domain": "economic", "excerpt": "Our analysis found that very few occupations see AI use across most of their associated tasks: only approximately 4% of jobs used AI for at least 75% of tasks.", "relation": "measured", "direction": "context", "facet": "Observed breadth of occupational-task use, distinct from AI's share of cognitive production", "why": "The analysis maps roughly one million Claude conversations to occupational tasks and finds broad within-occupation task coverage uncommon in that dataset. This is a specific usage baseline against which claims of economy-wide cognitive dominance would require much stronger, time- or value-weighted measurement.", "doesNotEstablish": "The dataset covers Claude.ai Free and Pro conversations, not a representative leading economy or all AI systems. Task mentions are not autonomous task completion, labor hours or output value. The 4% statistic cannot establish that AI supplies either less or more than half of cognitive labor.", "metric": { "value": 4, "unit": "percent of occupations", "coverage": "Approximate share with observed Claude use across at least 75% of associated tasks in the initial Economic Index dataset", "evidence": "Our analysis found that very few occupations see AI use across most of their associated tasks: only approximately 4% of jobs used AI for at least 75% of tasks." }, "reuseFamily": "observed-llm-occupational-task-coverage", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2032-1": [ { "id": "2032-1", "sourceId": "ref-bb9bfbd05aa4", "predictionText": "Advanced robots can perform roughly one third of economically valuable physical tasks", "predictionSha256": "94105438e417051417c28855e4819d73a9c652155b5f585da6ba793219e4fcda", "predictionTextSha256": "d3329d3d3040a1fe8019f47650c63fe80a63172c2bde32410c8538f27c139729", "domain": "technology", "excerpt": "Based on extensive surveys asking \"what do you want robots to do for you?\", we present 1,000 household activities that people actually spend time on and want help with.", "relation": "precursor", "direction": "context", "facet": "Human-needs-grounded benchmark of long-horizon physical activities", "why": "BEHAVIOR-1K supplies a specific evaluation framework spanning 1,000 household activities rather than a handful of selected robot demonstrations. Its survey grounding is relevant to measuring useful physical-task breadth.", "doesNotEstablish": "The benchmark's existence is not successful robot performance. Its simulated household activities are neither an employment-weighted denominator nor a representative census of all economically valuable physical tasks, and it establishes no one-third success threshold.", "metric": { "value": 1000, "unit": "household activities", "coverage": "Survey-grounded BEHAVIOR-1K simulation benchmark; not tasks already mastered by robots", "evidence": "we present 1,000 household activities that people actually spend time on and want help with." }, "reuseFamily": "human-relevant-physical-task-evaluation", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2032-2": [ { "id": "2032-2", "sourceId": "ref-569faca09ce6", "predictionText": "AI-driven real GDP growth reaches about 50% in at least one leading economy", "predictionSha256": "7f5cfa039d29a3a633cee375590ebf3c8df581a5db63a1b810128762640bfe48", "predictionTextSha256": "a938390495b92376ba670fd9000b6ae875d8f7f8be383cc1d9581298bf38d5f1", "domain": "economic", "excerpt": "Consequently, predicted TFP gains over the next 10 years are even more modest and are predicted to be less than 0.53%.", "relation": "constraint", "direction": "challenges", "facet": "Task-weighted aggregation limits on translating local AI productivity gains into macroeconomic growth", "why": "Acemoglu's task-based framework links aggregate gains to the fraction of affected tasks and their cost savings. Its conditional estimate illustrates why strong performance on selected tasks need not produce explosive macroeconomic growth; 50% real GDP growth would require mechanisms far beyond this calibration.", "doesNotEstablish": "The less-than-0.53% number is a modeled cumulative ten-year TFP gain, not observed growth or an annual GDP rate. It is not a universal upper bound and does not rule out future technologies, new tasks or scientific innovations absent from the model.", "metric": { "value": 0.53, "unit": "percent cumulative TFP gain", "coverage": "Author's conditional ten-year estimate after adjusting for harder-to-learn tasks; a projection, not an observed outcome", "operator": "<", "evidence": "Consequently, predicted TFP gains over the next 10 years are even more modest and are predicted to be less than 0.53%." }, "reuseFamily": "task-cost-savings-to-aggregate-productivity-constraint", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2032-3": [ { "id": "2032-3", "sourceId": "ref-e9b8dee2be94", "predictionText": "Capital floods into mines, motors, actuators, fabs and factories as robotics becomes the binding bottleneck", "predictionSha256": "5bb23b622b0935904a2ff244fa32eeb2b74dd86430e16228f49fe351786fdcff", "predictionTextSha256": "b686046133c7e1583033869990c2aa71ba777134a64dafa329f7a61084c973f3", "domain": "economic", "excerpt": "Investment momentum in critical mineral development weakened in 2024, with spending rising by just 5%, down from 14% in 2023.", "relation": "counterevidence", "direction": "challenges", "facet": "Weak mineral-investment response despite expectations of growing demand", "why": "The IEA reports slowing mineral investment and explains that low prices and uncertainty weaken investment incentives. This challenges an automatic inference from anticipated hardware demand to a rapid flood of upstream capital.", "doesNotEstablish": "The observation predates 2032 and does not rule out a later investment boom. It concerns critical minerals, not measured robotics-driven capital flows into motors, actuators, fabs or factories, and does not identify robotics as the binding bottleneck.", "metric": { "value": 5, "unit": "percent annual nominal spending growth", "coverage": "Critical-mineral development investment in 2024; the report separately puts real growth at 2%", "evidence": "Investment momentum in critical mineral development weakened in 2024, with spending rising by just 5%, down from 14% in 2023." }, "reuseFamily": "critical-mineral-investment-and-supply-response", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2032-4": [ { "id": "2032-4", "sourceId": "ref-c0a0d9ec675d", "predictionText": "At least one major jurisdiction caps or auctions permits for frontier compute and large-scale robot production", "predictionSha256": "5acfd0745d2a2484e6eb5593f61c3f1dfb19ebe03b33243108698ac3bb66488a", "predictionTextSha256": "a9ae75611449cdc964e71b20c13e7facb7177450463c74829b8c354fcb99e75a", "domain": "governance", "excerpt": "expands the Regional Stability (RS) license requirements and amends the RS licensing policy to adopt an additional case-by-case license review policy", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Existing governmental licensing of access to advanced-computing hardware", "why": "The rule establishes concrete advanced-computing export controls, destination restrictions and licensing review mechanisms. It demonstrates an administrative permission mechanism over strategically important compute hardware.", "doesNotEstablish": "Export licensing is not a domestic aggregate frontier-compute cap or an auction. This historical rule does not cover large-scale robot-production permits or establish the forecast combined regime.", "metric": null, "reuseFamily": "advanced-computing-export-licensing", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2032-5": [ { "id": "2032-5", "sourceId": "ref-ce8d6868fef6", "predictionText": "Tax systems begin shifting materially from human income toward compute, robot and automated-capital rents", "predictionSha256": "309066f9b334586afaf2fee799ebeb0c90ba5563f6081bb91d80017019df20b3", "predictionTextSha256": "97ff17599ecfb0b35d0e05530f63c1891997f8d15d1ef6d80df08a5a7be51c7b", "domain": "governance", "excerpt": "We find that it is optimal to tax robots while the current generations of routine workers, who can no longer move to non-routine occupations, are active in the labor force.", "relation": "policy", "direction": "context", "facet": "Economic rationale and limits for taxing automation during labour-market transition", "why": "The quantitative model explicitly evaluates robot taxation in response to automation-driven inequality. Importantly, its optimal robot tax is transitional and becomes zero after the affected routine-worker generations retire.", "doesNotEstablish": "This is a model-based policy analysis, not an enacted tax or measured revenue shift. It does not establish compute-rent taxes, durable automated-capital taxation, or material displacement of human-income taxation by 2032.", "metric": null, "reuseFamily": "transitional-robot-tax-policy-analysis", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2033-0": [ { "id": "2033-0", "sourceId": "ref-57bd7c78c8c4", "predictionText": "AI-and-robot labor contributes at least half of economic output in a leading economy", "predictionSha256": "563d57876c519176b23d742b5d4b1ecfdd3bfcc29e0ac9b7793fbb060bbd5c97", "predictionTextSha256": "327d762d174c8f59d6ae97c9977619725381dadc6c01d89950a7269485d7f0b1", "domain": "economic", "excerpt": "We calculate that the increased use of robots raised countries' average growth rates by about 0.37 percentage points. We also find that robots increased both wages and total factor productivity.", "relation": "measured", "direction": "supports-prerequisite", "facet": "Empirically estimated contribution of industrial robotics to value added and productivity", "why": "Using industry panels across 17 countries from 1993–2007, the authors estimate positive robot effects on value added and productivity. This supplies empirical evidence for the mechanism that robot capital can contribute to economic production, rather than treating robot installations as output automatically.", "doesNotEstablish": "An estimated 0.37-percentage-point contribution to average growth is not a 0.37% output share and cannot be extrapolated to half of GDP. The historical study does not measure combined modern AI-and-robot labor or identify a leading economy with majority machine-produced output.", "metric": { "value": 0.37, "unit": "percentage points of average growth", "coverage": "Authors' estimated contribution from increased industrial-robot use in a 17-country industry panel, 1993–2007", "evidence": "We calculate that the increased use of robots raised countries' average growth rates by about 0.37 percentage points. We also find that robots increased both wages and total factor productivity." }, "reuseFamily": "industrial-robot-value-added-growth-contribution", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2033-1": [ { "id": "2033-1", "sourceId": "ref-1b81e5c02a10", "predictionText": "A recurring citizen's dividend funded by AI, compute or robot rents launches in at least one major economy", "predictionSha256": "da284d321355a52e2c5cd009d2728f1c33bc88ff55f832d50c4b5172faf0a668", "predictionTextSha256": "e303d8a4ec7e49dfe799ea0f503a643ac6095ffb245f06a0ac16d10278521803", "domain": "economic", "excerpt": "The State Constitution directs that at least 25% of Alaska’s mineral royalties be deposited into the Principal. Alaska statutes mandate 50% for leases after 1979.", "relation": "policy", "direction": "supports-prerequisite", "facet": "An enacted mechanism converting public resource rents into a fund supporting recurring resident dividends", "why": "Alaska constitutionally captures mineral royalties in a protected fund and uses realized investment earnings to support annual dividends. This is a concrete institutional analogue for collecting scarce-resource rents and distributing a recurring public benefit.", "doesNotEstablish": "The rents are mineral royalties, not AI, compute or robot rents. Alaska is a subnational jurisdiction, not a major national economy. The source does not establish an AI-dividend launch, its tax base, coverage or adequacy.", "metric": { "value": 25, "unit": "percent minimum constitutional royalty deposit", "coverage": "Alaska mineral royalties directed to the Permanent Fund principal; statutory deposits are higher for specified later leases", "evidence": "The State Constitution directs that at least 25% of Alaska’s mineral royalties be deposited into the Principal. Alaska statutes mandate 50% for leases after 1979." }, "reuseFamily": "alaska-rent-funded-statutory-resident-dividend", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2033-2": [ { "id": "2033-2", "sourceId": "ref-5b0fdfb5ea56", "predictionText": "Cross-border distribution of AI-generated wealth becomes a serious geopolitical negotiation", "predictionSha256": "490820e76713a07a348e78a7b31ad8be94264c5c58b2f54f7b8f450ddffe6bf7", "predictionTextSha256": "89b05f7882f0fd53e98a1eabf03ba56d2cc4c955aab31ef326d106493fb84364", "domain": "geopolitical", "excerpt": "We recommend the creation of a global fund for AI to put a floor under the AI divide.", "relation": "policy", "direction": "supports-prerequisite", "facet": "A specific international proposal to pool and distribute financial and in-kind AI resources", "why": "Recommendation 5, page 17, proposes a global fund receiving public and private financial and in-kind contributions and distributing access to compute, models and related capacity. This is a concrete cross-border benefit-sharing proposal.", "doesNotEstablish": "The proposal is not evidence that the fund exists or that governments are negotiating redistribution of AI profits. It creates no demonstrated dividend or compensation right, and capacity assistance is narrower than distribution of AI-generated wealth.", "metric": null, "reuseFamily": "international-ai-capacity-fund-benefit-sharing", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2033-3": [ { "id": "2033-3", "sourceId": "ref-1ed19ea62985", "predictionText": "Cheap AI persuasion triggers capability limits, disclosure rules or taxes on targeted influence", "predictionSha256": "59701b2421e0244e5f997fd80dc4730f655ca54fcfeca376401fcf2783e1d1a4", "predictionTextSha256": "8fc3e2629977ddd02a629cc9c073e39069df687371c25d9c35231a2a33641b33", "domain": "governance", "excerpt": "the data subject has provided explicit consent within the meaning of Regulations (EU) 2016/679 and (EU) 2018/1725 to the processing of personal data separately for the purpose of political advertising;", "relation": "policy", "direction": "supports-prerequisite", "facet": "Binding restrictions on personal-data targeting for political advertising", "why": "Article 18 imposes specific consent and data-source conditions on online political-advertising targeting and excludes profiling using special-category data. The regulation supplies an actual legal mechanism constraining targeted influence.", "doesNotEstablish": "It does not show that cheap AI persuasion caused these rules, that AI persuasion crossed a capability threshold, or that influence taxes or general persuasion-model capability caps exist.", "metric": null, "reuseFamily": "political-advertising-personal-data-targeting-controls", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2033-4": [ { "id": "2033-4", "sourceId": "ref-5007e8060d43", "predictionText": "Personal truth-seeking AI advisors begin replacing one-size-fits-all feeds and search interfaces", "predictionSha256": "202b1559d61a8967c33b3ac1eca441a05dc5ed1f293c219d5b2405d2adc1029c", "predictionTextSha256": "c4c2681f80d71157b1d4bdb67224b7642fb36ba6029bc5e5504eeb0586adfb4f", "domain": "individual", "excerpt": "You can ask anything on your mind and get a helpful AI-powered response with the ability to go further with follow-up questions and helpful web links.", "relation": "trial", "direction": "supports-prerequisite", "facet": "A conversational, query-specific and source-linked alternative inside a mainstream search interface", "why": "Google introduced an opt-in AI Mode experiment that answers nuanced questions, supports follow-ups and retrieves across multiple searches. This is an actual interface trial moving beyond a fixed search-results interaction toward individualized inquiry.", "doesNotEstablish": "An announced experiment is not evidence that personal advisors have displaced search or feeds at population scale. Google explicitly acknowledges factuality and persona/opinion errors. The source does not establish durable personalization, independence, truth-seeking reliability or replacement of social feeds.", "metric": null, "reuseFamily": "conversational-source-linked-search-interface-trial", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2033-5": [ { "id": "2033-5", "sourceId": "ref-ee87371859e6", "predictionText": "Unregulated AI-and-robot economies demonstrate subannual doubling potential", "predictionSha256": "26c4ed7be4d7c9c4b65e8c033945c04b85b8bd91f70f5642a97c5ced52039838", "predictionTextSha256": "604e2ef25f3965f85bfcdccf7c64f1dde5980ba6782bbd8728154d9c1dfdabc9", "domain": "economic", "excerpt": "One theme that emerges is based on Baumol’s “cost disease” insight: growth may be constrained not by what we are good at but rather by what is essential and yet hard to improve.", "relation": "constraint", "direction": "challenges", "facet": "Essential-input bottlenecks in theoretical AI-driven growth and singularity scenarios", "why": "The authors explicitly analyze AI, automation, growth and singularity questions. Their bottleneck mechanism explains why rapid improvements in automated activities need not produce subannual economy-wide doubling: indispensable activities that improve slowly can remain binding even without regulatory restrictions.", "doesNotEstablish": "This is theoretical research, not an observed unregulated AI-and-robot economy or a measured doubling time. It does not provide a universal impossibility theorem. Substitution, further automation and new technologies could change which inputs bind.", "metric": null, "reuseFamily": "essential-input-bottlenecks-in-ai-macro-growth", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2033-6": [ { "id": "2033-6", "sourceId": "ref-20c7e85747cb", "predictionText": "Governments fund universal-scale biodefense, rapid vaccines and continuous pathogen monitoring", "predictionSha256": "e8a3b541a45d856809aedcba1dca01196cc0ba338ebf462e68c23f95a463197b", "predictionTextSha256": "58cc45ac11100a16837e849d1f7d433c8d88cd5ee5ae9c6bc8c4764cb7d8f0ba", "domain": "governance", "excerpt": "CDC's wastewater monitoring program, also known as the National Wastewater Surveillance System (NWSS), provides the public health infrastructure to monitor infectious diseases through wastewater across the country.", "relation": "deployment", "direction": "supports-prerequisite", "facet": "Government-supported nationwide wastewater pathogen surveillance infrastructure", "why": "CDC describes an operating surveillance system with state and local data funded through its ELC cooperative agreement, a national wastewater testing contract, and academic-partner inputs. These are concrete public funding and implementation mechanisms for pathogen monitoring.", "doesNotEstablish": "Participating wastewater sites do not imply universal population coverage or continuous real-time monitoring. This source does not establish universal-scale biodefense, rapid-vaccine capability, coverage of all pathogens, or commitments through 2033.", "metric": null, "reuseFamily": "publicly-funded-wastewater-pathogen-surveillance", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2034-0": [ { "id": "2034-0", "sourceId": "ref-78446ff77822", "predictionText": "AI automates a majority of cognitive work in semiconductor R&D and production engineering", "predictionSha256": "cd0d3e980977f23eb109773a359c9c7757d62aa8bc93d940e4d781d3c4e99219", "predictionTextSha256": "2a9dc2627c0ae37f9a409786185b232cf329dbbd50ff061f4513133cf332acb8", "domain": "technology", "excerpt": "The method has been used to design superhuman chip layouts in the last three generations of Google’s custom AI accelerator, the Tensor Processing Unit (TPU).", "relation": "deployment", "direction": "supports-prerequisite", "facet": "Research-team-reported production use of reinforcement learning for chip floorplanning", "why": "Google reports AlphaChip-generated layouts in multiple TPU generations, including designs used in production hardware. This is a concrete instance of AI performing a semiconductor engineering subtask beyond a research-only benchmark.", "doesNotEstablish": "The source's performance language is the developer's characterization. Floorplanning is only one part of semiconductor work; the report does not measure labor-hour shares across architecture, verification, process development, fabrication or production engineering, much less majority automation by 2034.", "metric": { "value": 3, "unit": "TPU generations", "coverage": "Google-reported use of AlphaChip layouts as described in the September 2024 article; not the share of engineering work automated", "evidence": "in the last three generations of Google’s custom AI accelerator, the Tensor Processing Unit (TPU)." }, "reuseFamily": "ai-semiconductor-floorplanning-deployment", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2034-1": [ { "id": "2034-1", "sourceId": "ref-a06d967a4287", "predictionText": "Continuously running AI agents form a virtual workforce of at least 100 million copies", "predictionSha256": "47511732587904994b998c66bf83a3586939d424b9592f103ed48170a335dca9", "predictionTextSha256": "d9de1c3021ed1fe12f2eba4241a290a65d2b09676199335aa00f8411ef691b62", "domain": "economic", "excerpt": "Introduced Agentic Work Units (“AWUs”) to measures tasks accomplished by an AI Agent, with 2.4 billion AWUs delivered to date across Agentforce and Slack, growing 57% quarter-over-quarter (“Q/Q”)", "relation": "deployment", "direction": "supports-prerequisite", "facet": "Production-scale enterprise agent task execution", "why": "Salesforce reports 2.4 billion cumulative Agentic Work Units across Agentforce and Slack. This demonstrates a company-reported deployed agent-work baseline at substantial transaction scale, an operational prerequisite for a larger virtual workforce.", "doesNotEstablish": "Cumulative AWUs are vendor-defined task units, not distinct or concurrent agents. They do not establish 100 million copies, continuous operation, worker-equivalent output, reliability or utilization. Repeated actions by a small fleet could generate a large cumulative count.", "metric": { "value": 2400000000, "unit": "cumulative Agentic Work Units", "coverage": "Company-reported tasks across Agentforce and Slack delivered to date in Salesforce's February 2026 release", "evidence": "Introduced Agentic Work Units (“AWUs”) to measures tasks accomplished by an AI Agent, with 2.4 billion AWUs delivered to date across Agentforce and Slack, growing 57% quarter-over-quarter (“Q/Q”)" }, "reuseFamily": "salesforce-commercial-agent-work-deployment", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2034-2": [ { "id": "2034-2", "sourceId": "ref-7a179a48aad8", "predictionText": "Global AI compute reaches multi-terawatt scale and billions of H100-equivalents", "predictionSha256": "0f7987c833ce0901a5586aef0403093ac22ac9f2b25a09752ffa77b9435b3fa3", "predictionTextSha256": "8c7c8eef7a2d332f7bfd372ad8d0876eaf84553cb9e3e7a4abdeb2be00c8d32e", "domain": "economic", "excerpt": "Uncertainties widen further after 2030, but our Base Case sees global data centre electricity consumption rising to around 1 200 TWh by 2035.", "relation": "counterevidence", "direction": "challenges", "facet": "A substantially smaller official baseline projection for global datacenter electricity consumption", "why": "The IEA's 2035 base case for all datacenters represents much less average electrical demand than continuously operating multi-terawatt infrastructure. It provides a credible alternative scale trajectory that the forecast would have to substantially exceed.", "doesNotEstablish": "This is a scenario projection, not an upper physical bound. Annual electricity consumption differs from nameplate capacity, and the report includes non-AI datacenters. It does not define H100-equivalents or exclude unusually rapid deployment, efficiency changes, or a different utilization rate.", "metric": { "value": 1200, "unit": "TWh per year", "coverage": "IEA base-case projected global datacenter electricity consumption in 2035, including non-AI workloads", "evidence": "our Base Case sees global data centre electricity consumption rising to around 1 200 TWh by 2035." }, "reuseFamily": "datacenter-capital-and-electricity-buildout", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2034-3": [ { "id": "2034-3", "sourceId": "ref-44a63fa0e787", "predictionText": "Managed branch: major powers adopt compute caps or mutually assured compute-destruction provisions", "predictionSha256": "b5ce6f56abeaeaa1c28b2ed76be7797621c555be637655327ce627c9ba4f0d52", "predictionTextSha256": "1c79b83c2da4e5642ac980752f5b87cfa953e60d165156b4a3b1c5468ded7348", "domain": "geopolitical", "excerpt": "If these conditions are not met, it would block the chip from operating.", "relation": "feasibility", "direction": "supports-prerequisite", "facet": "Proposed hardware operating licences capable of denying compute operation", "why": "The report proposes hardened security modules that condition chip operation on valid firmware, software and, where applicable, a renewed operating licence. This is a specific technical enforcement precursor to restrictions on compute use.", "doesNotEstablish": "The design is a proposal requiring security hardening, not an adopted great-power agreement. Revoking an operating licence is not physical destruction, and the source establishes no mutually assured destruction provision or aggregate compute cap.", "metric": null, "reuseFamily": "hardware-enforced-adaptive-compute-licensing", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2034-4": [ { "id": "2034-4", "sourceId": "ref-2cd5a560d120", "predictionText": "Large datacenters expand into neutral jurisdictions, international waters or other auditable locations", "predictionSha256": "c477d84f46b91d289c81957d955b2f7a2ee8d7295472a4683b648b832e0ddb96", "predictionTextSha256": "b73b03217cc012fbf56eda644e55927a1780b757c1227057d1a4da857fc7549a", "domain": "geopolitical", "excerpt": "Microsoft’s Project Natick team deployed the Northern Isles datacenter 117 feet deep to the seafloor in spring 2018.", "relation": "feasibility", "direction": "supports-prerequisite", "facet": "Reported multiyear operation of a sealed offshore datacenter", "why": "Microsoft describes an 864-server vessel operated and monitored for two years off the Orkney Islands. This is a concrete engineering precursor for placing computing infrastructure outside conventional land-based buildings.", "doesNotEstablish": "The Scottish coastal experiment was not an international-waters or neutral-jurisdiction deployment. It does not demonstrate hyperscale commercial expansion, international auditing arrangements, or a governance motive for offshore placement.", "metric": { "value": 864, "unit": "servers", "coverage": "Microsoft-reported Northern Isles proof-of-concept vessel off Orkney; not a hyperscale or extraterritorial facility", "evidence": "steel tube that encased the Northern Isles’ 864 servers and related cooling system infrastructure." }, "reuseFamily": "offshore-sealed-datacenter-feasibility", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2034-5": [ { "id": "2034-5", "sourceId": "ref-0bcbc033eb9e", "predictionText": "Treaties constrain the use of frontier AI for military R&D and autonomous strategic weapons", "predictionSha256": "a53704eea0e7aad9c629543a0b10780d377b106b5d750239db3269b8db05c76e", "predictionTextSha256": "22c521b2543c70fcc761dc5ff227fd12d037c4179ad42bbb1794251cf824b601", "domain": "governance", "excerpt": "New Zealand is committed to working towards effective, multilaterally agreed rules and limits on autonomous weapon systems (AWS). We continue to seek binding international prohibitions and regulation of AWS.", "relation": "policy", "direction": "supports-prerequisite", "facet": "Government pursuit of legally binding multilateral autonomous-weapons rules", "why": "MFAT identifies active engagement in the CCW expert process and explicitly distinguishes voluntary military-AI initiatives from the binding rules it seeks. This is a real governmental negotiating objective with an existing institutional forum.", "doesNotEstablish": "Seeking binding rules is not an enacted treaty. The source does not establish constraints specifically on frontier military R&D, autonomous strategic weapons, or acceptance of such limits by all major powers.", "metric": null, "reuseFamily": "autonomous-weapons-binding-rule-diplomacy", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2035-0": [ { "id": "2035-0", "sourceId": "ref-05665ed86ec7", "predictionText": "Managed branch: the strongest AIs reach top-human-expert capability across essentially every cognitive field", "predictionSha256": "2513152df3c74eaaa3a3548cb56aba89d948cc44fca541f89bfe8fc95ae9682f", "predictionTextSha256": "d2c180f7fd3d7f0d6ff927f0e129f972c6474b48c2dfe6e19af5737980b726bf", "domain": "technology", "excerpt": "All frontier models achieve low accuracy on Humanity's Last Exam, highlighting significant room for improvement in narrowing the gap between current LLMs and expert-level academic capabilities on closed-ended questions.", "relation": "constraint", "direction": "context", "facet": "A broad expert-academic benchmark reveals remaining gaps in its displayed model snapshot", "why": "HLE's 2,500 questions across more than 100 subjects operationalize one slice of the breadth demanded by the forecast. Its displayed results supply a concrete reference against which claims of universal expert competence can be compared.", "doesNotEstablish": "The quoted statement and displayed leaderboard are a historical project snapshot, not an exhaustive September 2026 frontier assessment. Closed-ended academic questions do not cover essentially every cognitive field, expert practical work, or a managed governance branch; the result does not determine 2035 capabilities.", "metric": { "value": 38.3, "unit": "percent accuracy", "coverage": "Gemini 3 Pro in the project's displayed HLE table at retrieval; no claim that this is the latest model, latest evaluation, or best currently achievable score.", "evidence": "Model Accuracy (%) ↑ Calibration Error (%) ↓ Gemini 3 Pro 38.3 57.2" }, "reuseFamily": "cross-domain-closed-ended-expert-evaluation", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2035-1": [ { "id": "2035-1", "sourceId": "ref-06c10deb3670", "predictionText": "Managed branch: frontier capabilities pause near top-expert level because control no longer scales safely", "predictionSha256": "604b8a1e5bcd06f6ff98c47b412ecef6f9bf30ad8429e729ac52872eb402311a", "predictionTextSha256": "6c4ac58c1eeb7867cdc1806fc2b8a250c7b744cc529b6f5a02ef2fc7e04f88e8", "domain": "governance", "excerpt": "In particular, we cannot unilaterally and unconditionally commit to staying in line with the industry-wide recommendations in the right column.", "relation": "counterevidence", "direction": "challenges", "facet": "Limits of extrapolating an old company safety pledge into a dependable capability pause", "why": "As of the September 5, 2026 review: The official policy index lists version 3.4, effective July 8, 2026, as current. Its introduction explicitly changes the earlier RSP approach: company plans are separated from industry-wide recommendations that Anthropic cannot commit to following unilaterally. This materially qualifies reliance on the October 2024 pledge as a present unconditional development gate.", "doesNotEstablish": "This challenges that specific institutional precursor, not the entire future managed-branch forecast. Other safeguards and competitor-contingent commitments remain, and discretionary pauses remain possible. A voluntary company policy neither demonstrates control failure nor establishes whether governments or industry will pause near top-expert capability. July 8 is the policy's effective date, not an independently established publication timestamp.", "metric": null, "reuseFamily": "current-responsible-scaling-policy-governance", "reviewedAt": "2026-09-05T09:55:12.063Z", "reviewedBy": "Agent review under user authorization" } ], "2035-2": [ { "id": "2035-2", "sourceId": "ref-d9afef730d39", "predictionText": "AI and robots perform 85% or more of economically valuable labor in at least one leading economy", "predictionSha256": "4cf4dd077222bb6e5ae08dfec3f7bcda2eb446ab86e6897d77af02554417ee3a", "predictionTextSha256": "eb4f63038aa0e6681e1cbe06dfdf2de2da5e4d675a176f9783da19002e55b0b6", "domain": "economic", "excerpt": "adoption depends not only on technical feasibility (i.e., AI's absolute advantage measured by exposure) but on profitability (i.e., AI's comparative (dis)advantage relative to a specific worker)", "relation": "constraint", "direction": "challenges", "facet": "The profitability gap between technically exposed work and actual workplace AI adoption", "why": "This newer primary working paper links the representative German DiWaBe employee survey to worker and establishment data. It finds a weak relationship between exposure measures and worker-reported AI use, and models adoption using relative productivity, user costs and wages. It provides a concrete economic constraint on extrapolating technical capability into an 85% labor-performance share.", "doesNotEstablish": "This is an author working paper, not a peer-reviewed causal forecast. Its adoption index and explained variation are not the fraction of work automated, an AI-plus-robot labor share, or a national 85% outcome. It does not reproduce the inaccessible MIT computer-vision paper's 23% estimate; that result is not imported into this different study.", "metric": { "value": 60, "operator": "~", "unit": "percent of cross-occupation adoption variation", "coverage": "Almost 60% explained by the study's occupation-level index in its observed German AI-adoption data; not automated labor share", "evidence": "The resulting occupation-level index accounts for almost 60% of cross-occupation variation in observed AI adoption, compared to 14% for an exposure-only model." }, "reuseFamily": "comparative-advantage-ai-adoption-economics", "reviewedAt": "2026-09-05T09:53:43.793Z", "reviewedBy": "Agent review under user authorization" } ], "2035-3": [ { "id": "2035-3", "sourceId": "ref-1b81e5c02a10", "predictionText": "Universal high income or an AI dividend becomes a permanent institution in multiple major economies", "predictionSha256": "ccf7f95f246547e6269f935a3581ad7441bfcee35be062799961dabe1aeddd26", "predictionTextSha256": "c2df835660ccbf1724c2a99e5ac7e2502d10ecd3e62774de72ff4e54c0ab32c6", "domain": "economic", "excerpt": "To date, the primary use of the ERA has been to pay an annual dividend to the citizens of Alaska.", "relation": "policy", "direction": "supports-prerequisite", "facet": "Durable legal and investment architecture for recurring public dividends", "why": "Alaska's protected principal, statutory earnings account and rules-based withdrawals show how a recurring citizen payment can be embedded in a lasting public institution rather than run solely as a short pilot. The structure also exposes sustainability and appropriation requirements.", "doesNotEstablish": "This is one subnational mineral-rent institution, not multiple major economies, an AI-funded dividend or universal high income. Payments remain subject to eligibility and appropriations; the architecture does not guarantee a fixed, sufficient or legally unchangeable income.", "metric": null, "reuseFamily": "alaska-rent-funded-statutory-resident-dividend", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2035-4": [ { "id": "2035-4", "sourceId": "ref-0e4704ac5d99", "predictionText": "AI persuasion exceeds the best human persuaders and is subject to strict international controls", "predictionSha256": "b0de7e048369a5de4844f0c509f4e5818613f8be497a1e4f2ac792ab53a67775", "predictionTextSha256": "367114f68e9efc9fd705a0b951afdb6e46ae5837e5dc39d8bb68137af9f4c5e0", "domain": "social", "excerpt": "In debate pairs where AI and humans were not equally persuasive, GPT-4 with personalization was more persuasive 64.4% of the time", "relation": "trial", "direction": "supports-prerequisite", "facet": "Controlled experimental comparison of an LLM's conversational persuasion against human opponents", "why": "The preregistered 900-participant experiment provides direct human-comparator evidence. The current corrected text reports 81.2% higher odds of post-debate agreement for personalized GPT-4 versus the human baseline. The authors interpret the result as a 64.4% conditional probability that AI is more persuasive among pairs where AI and humans are not equally persuasive, not 64.4% greater persuasiveness.", "doesNotEstablish": "Human opponents were recruited participants, not the best professional persuaders. Structured short debates do not establish real-world dominance or strict international controls. Personalized versus non-personalized GPT-4 was not significantly different at the conventional threshold, so personalization alone is not proven to cause the advantage.", "metric": { "value": 64.4, "unit": "percent conditional probability", "coverage": "Authors' interpretation of personalized GPT-4 being more persuasive than human opponents, conditional on unequal persuasiveness; not an unconditional persuasion success rate", "evidence": "In debate pairs where AI and humans were not equally persuasive, GPT-4 with personalization was more persuasive 64.4% of the time" }, "reuseFamily": "controlled-human-versus-llm-conversational-persuasion", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2035-5": [ { "id": "2035-5", "sourceId": "ref-5cea71b71d6e", "predictionText": "AI welfare, compensation and legal status enter mainstream law and corporate governance", "predictionSha256": "82b805903f4db59a9848740de20970bf499d7a3410fb3c5212c0cc876cf954a7", "predictionTextSha256": "e0babe9adf96bc55b1ecfee9e85fdfc7e9275f09dd89a0c67f8f055528c98ca5", "domain": "social", "excerpt": "To that end, we recently started a research program to investigate, and prepare to navigate, model welfare.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "A frontier AI company's explicit institutional research program on model welfare", "why": "Anthropic reports starting a model-welfare program addressing possible moral consideration, preferences, distress and practical interventions. This is a concrete corporate organizational precursor to welfare governance rather than merely an outside philosophical proposal.", "doesNotEstablish": "A research program is not mainstream law, legal personhood, compensation rights or an enacted corporate welfare regime. Anthropic explicitly states that consciousness and morally relevant experience remain scientifically uncertain; the source does not establish that current models are conscious.", "metric": null, "reuseFamily": "corporate-model-welfare-research-institutionalization", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2035-6": [ { "id": "2035-6", "sourceId": "ref-fdc10e3226e6", "predictionText": "Mechanistic interpretability becomes a practical tool for detecting deception and tracing model decisions", "predictionSha256": "fceed6346fd4fe3148bb61ebd76da54c7ae523f2cf5c43042040fae6883de631", "predictionTextSha256": "26ac6aca65da5cc4d610776cf36522009f2342094b9c7f61ff90914b2950263a", "domain": "technology", "excerpt": "We are able to “catch it in the act” as it makes up its fake reasoning, providing a proof of concept that our tools can be useful for flagging concerning mechanisms in models.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Circuit tracing identifies unfaithful mathematical rationalization rather than trusting the model's explanation", "why": "The researchers distinguish calculations supported by internal computational paths from plausible reasoning constructed around a supplied answer. This directly demonstrates a mechanism-level auditing use relevant to tracing decisions and detecting misleading explanations.", "doesNotEstablish": "Detecting fabricated reasoning in selected short prompts is not a validated detector of strategic deception in deployed agents. The work captures only part of computation, may introduce interpretive artifacts, and does not report general sensitivity, specificity, or scalable operational coverage.", "metric": null, "reuseFamily": "attribution-circuit-reasoning-faithfulness", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2036-0": [ { "id": "2036-0", "sourceId": "ref-08731f05d093", "predictionText": "The global economy runs at least 200 million frontier AI workers and 2 billion advanced robots", "predictionSha256": "1135811540bca0894badb1281f0cc6c86fa094be0b418e1af3b996c3c33e7bad", "predictionTextSha256": "64806bce359064d859aada39231621c5d33b4a966f9518b0ea39671733903285", "domain": "economic", "excerpt": "The total number of industrial robots in operational use worldwide was 4,664,000 units in 2024 – an increase of 9% compared to the previous year.", "relation": "measured", "direction": "context", "facet": "Updated historical global operational stock of industrial robots", "why": "IFR's first-party World Robotics 2025 release provides a newer global installed-stock baseline for 2024. It explicitly distinguishes robots in operational use from annual installations, preserving the relevant stock measure for comparison with the forecast's two-billion-robot scale.", "doesNotEstablish": "Industrial robots are not equivalent to the forecast's undefined advanced robots, and this is not an all-category robot census. The published figure should not be treated as unit-level precision. The release supplies no frontier-AI-worker count or deployment trajectory establishing either 2036 threshold.", "metric": { "value": 4664000, "unit": "operating industrial robots", "coverage": "IFR's worldwide industrial-robot operational stock for 2024, as published in the World Robotics 2025 release; not annual installations", "evidence": "The total number of industrial robots in operational use worldwide was 4,664,000 units in 2024" }, "reuseFamily": "industrial-robot-operational-stock", "reviewedAt": "2026-09-05T00:47:27.963Z", "reviewedBy": "Agent review under user authorization" } ], "2036-1": [ { "id": "2036-1", "sourceId": "ref-d0f5ca5d8a30", "predictionText": "AI and robots can perform about 95% of cognitive and physical tasks", "predictionSha256": "08496c13d63c04af873457e3677032d2cda8aecbc1d48def6ed11d108568faf3", "predictionTextSha256": "9ff4df10f165a99b23b79644d627673c6a82c65d26a276d19f56ec08dc58a24d", "domain": "technology", "excerpt": "In our tech report, we show that on average, Gemini Robotics more than doubles performance on a comprehensive generalization benchmark compared to other state-of-the-art vision-language-action models.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Research-team-reported improvement in robotic generalization", "why": "Generalization to new objects, instructions and environments is a specific prerequisite for broad physical-task coverage. The research announcement reports comparative benchmark gains for a vision-language-action model and describes operation across multiple embodiments.", "doesNotEstablish": "The comparison is reported by the model developer and is not a 95% task-coverage measurement. It supplies neither an economy-wide task denominator nor broad cognitive-work results, deployment reliability, or evidence that the target will be reached by 2036.", "metric": null, "reuseFamily": "embodied-robot-generality-and-dexterity", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2036-2": [ { "id": "2036-2", "sourceId": "ref-5efeb9ec3c3d", "predictionText": "Human employment falls below half of working-age adults in at least one leading economy", "predictionSha256": "dde07fe6d3a8972d1a61ca0a30e3e407d948bf808f81e26ba1c1e0a6e6f6f0e8", "predictionTextSha256": "e2d8ef6a8c3e15490c550a0d617b68728d91cbbb421c8fbb8938c14109beb33c", "domain": "economic", "excerpt": "The estimated UK employment rate decreased by 0.2 percentage points on the year, but increased by 0.1 percentage points, to 75.1%, in the latest quarter (April to June 2026)", "relation": "measured", "direction": "context", "facet": "Working-age employment-to-population baseline in a leading economy", "why": "The latest available ONS bulletin estimates that 75.1% of UK residents aged 16–64 were employed in April–June 2026. Relative to this point estimate, reaching below 50% would require a decline exceeding 25.1 percentage points, making the scale of the predicted labor-market change explicit.", "doesNotEstablish": "This baseline neither predicts a future decline nor attributes employment changes to AI. It covers ages 16–64 rather than every possible definition of working-age adults. The rate is an LFS sample estimate; ONS continues to caution against conclusions from short-term changes and recommends considering other labor-market indicators.", "metric": { "value": 75.1, "unit": "percent employed", "coverage": "UK population aged 16–64, April–June 2026, seasonally adjusted Labour Force Survey estimate in the August 2026 bulletin", "evidence": "The estimated UK employment rate decreased by 0.2 percentage points on the year, but increased by 0.1 percentage points, to 75.1%, in the latest quarter (April to June 2026)" }, "reuseFamily": "working-age-employment-population-baseline", "reviewedAt": "2026-09-05T01:02:29.2237142Z", "reviewedBy": "Agent review under user authorization" } ], "2036-3": [ { "id": "2036-3", "sourceId": "ref-ee87371859e6", "predictionText": "AI-driven annual GDP growth approaches 100% even under managed production caps", "predictionSha256": "3f30cfba366a88f2f53136ec3c62939239db830a5922e7aae53fa50994762ce1", "predictionTextSha256": "c5ff0f2652be43c019cf59a5ac1d28265626e7ab4a32b3aab5993e0b56c32c55", "domain": "economic", "excerpt": "One theme that emerges is based on Baumol’s “cost disease” insight: growth may be constrained not by what we are good at but rather by what is essential and yet hard to improve.", "relation": "constraint", "direction": "challenges", "facet": "Whether essential constrained inputs permit near-doubling aggregate output", "why": "The paper's essential-input bottleneck mechanism is directly relevant to growth under production limits. Near-100% GDP growth would require the unspecified caps not to bind indispensable inputs, or sufficiently strong substitution and productivity improvements to overcome those bottlenecks.", "doesNotEstablish": "The paper does not model this forecast's particular managed-cap regime or report observed 100% growth. It provides a conditional theoretical constraint, not a numerical universal ceiling. The forecast must specify which outputs or inputs are capped and how GDP is measured.", "metric": null, "reuseFamily": "essential-input-bottlenecks-in-ai-macro-growth", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2036-4": [ { "id": "2036-4", "sourceId": "ref-e9b8dee2be94", "predictionText": "Land, energy, raw materials and positional goods replace labor as the economy's dominant scarcities", "predictionSha256": "d5967a76594ffa468988821809c32191fc4b09ba5b8d0ae31b7c55dec3643ec3", "predictionTextSha256": "25c36afca5f035da21d053cf4c18456f7c95a734016443df989fc0d4415164b4", "domain": "economic", "excerpt": "the current mine project pipeline points to a potential 30% supply shortfall by 2035 due to declining ore grades, rising capital costs, limited resource discoveries and long lead times.", "relation": "constraint", "direction": "context", "facet": "Potential copper supply scarcity from geological and mine-development constraints", "why": "The IEA identifies a concrete raw-material constraint: the announced copper mine pipeline may not meet projected demand because of ore quality, capital costs, discoveries and development time. These mechanisms do not disappear merely because cognitive work becomes cheaper.", "doesNotEstablish": "The shortfall is conditional and projected, not a measured current deficit. It does not establish that raw materials dominate labor scarcity, assess land or positional goods, or rule out substitution, recycling and new investment.", "metric": { "value": 30, "unit": "percent potential supply shortfall", "coverage": "Projected copper supply gap by 2035 from the current mine project pipeline under the report's assumptions", "evidence": "the current mine project pipeline points to a potential 30% supply shortfall by 2035" }, "reuseFamily": "critical-mineral-investment-and-supply-response", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2036-5": [ { "id": "2036-5", "sourceId": "ref-5df33d3f8ba3", "predictionText": "Voting, civic participation and ownership replace labor income as ordinary people's main leverage", "predictionSha256": "f25ed0846896ff2750313f2617fa8b3e38e217a096a8d8f477e876dbd6f6d5b3", "predictionTextSha256": "636e10303fa767b7aad74b718b6a152ae8855e486246cef9645d37d34a140386", "domain": "governance", "excerpt": "You need to get the support of at least 1 million EU citizens, with thresholds (minimum numbers) in at least 7 EU countries", "relation": "precursor", "direction": "context", "facet": "An established non-employment channel for citizens to influence institutional agendas", "why": "The initiative process gives qualifying citizen campaigns meetings with the Commission, a European Parliament hearing and a formal response. It is a concrete civic-participation mechanism independent of employment or bargaining with an employer.", "doesNotEstablish": "The Commission is not obliged to propose legislation. The mechanism does not show civic power or ownership overtaking labour income as people's main leverage, and creates no AI-related ownership or dividend right.", "metric": { "value": 1000000, "unit": "supporting EU citizens", "coverage": "Initiative support threshold, additionally requiring national minimums in at least seven EU countries; not measured adoption or economic leverage.", "evidence": "You need to get the support of at least 1 million EU citizens, with thresholds (minimum numbers) in at least 7 EU countries" }, "reuseFamily": "citizen-initiative-institutional-participation", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2036-6": [ { "id": "2036-6", "sourceId": "ref-5f6f63c537e3", "predictionText": "Education and social institutions recenter on meaning, community, relationships and stewardship rather than employability", "predictionSha256": "e712acfde8e8f6f76215b049c5f8938f3d1a9a2e4ba00f22f28add8656598f48", "predictionTextSha256": "1e3d2394e029463d8e5478bf32222e3065bef2870e8e58bcee628c72a51e48e5", "domain": "individual", "excerpt": "and find their direction in a meaningful and responsible way, instead of simply receiving fixed instructions or directions from their teachers.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "An institutional education framework emphasizing agency, well-being, meaning and responsibility", "why": "The OECD framework explicitly places student agency and well-being alongside knowledge, skills, attitudes and values, with meaningful and responsible self-direction. It supplies a concrete institutional vocabulary for educational purposes broader than employability alone.", "doesNotEstablish": "The page describes an aspirational framework, not measured curriculum adoption or a completed shift in education and social institutions. It does not show employability being displaced, AI causing the change, or community and stewardship becoming society-wide organizing principles. The publication date is the webpage metadata date, not the framework's original launch date.", "metric": null, "reuseFamily": "education-agency-wellbeing-responsibility-framework", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2037-0": [ { "id": "2037-0", "sourceId": "ref-a130b8e6da20", "predictionText": "AI accelerates scientific progress by 10x to 1000x depending on the field", "predictionSha256": "b55eba2625a348d897e844b269c0f3129988a6094d0c8e04062e7aa24b2c5351", "predictionTextSha256": "24540dee858c6786848e3e0dc8ea934807ea377a436f9e1709430dba4ef21172", "domain": "technology", "excerpt": "Subsequent experiments validated these proposals, confirming that the suggested drugs inhibit tumor viability at clinically relevant concentrations in multiple AML cell lines.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "AI-generated biomedical hypotheses lead to experimentally validated cell-line results", "why": "The co-scientist's acute myeloid leukemia repurposing suggestions were followed by laboratory tests, demonstrating a link from AI hypothesis generation to empirical scientific work. That is a substantive prerequisite for accelerating discovery rather than merely producing plausible research prose.", "doesNotEstablish": "Cell-line validation does not establish clinical benefit or a tenfold-to-thousandfold field-level increase in scientific progress. The study involved expert guidance and selected applications, and did not measure a controlled end-to-end research speedup across fields.", "metric": null, "reuseFamily": "ai-hypothesis-generation-with-laboratory-validation", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2037-1": [ { "id": "2037-1", "sourceId": "ref-273d762a7c91", "predictionText": "AI-driven research delivers major disease cures and abundant low-cost clean energy", "predictionSha256": "a087bd14f0ac193a7f878630252307e7b2daf507084eb02ae4b08b808f5765ea", "predictionTextSha256": "c7030faab074d1efdca69b648a36c5e38b8955476cba41e9b118bdccf1894e88", "domain": "technology", "excerpt": "Using a learning architecture that combines deep RL and a simulated environment, we produced controllers that can both keep the plasma steady and be used to accurately sculpt it into different shapes.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Experimentally tested AI control of fusion-research plasma", "why": "The research team describes training controllers in simulation and validating them on EPFL's real TCV tokamak. A single neural network controls the magnetic coils, demonstrating a concrete way AI can assist fusion experimentation.", "doesNotEstablish": "Plasma control is not net-positive electricity generation, a commercial fusion plant, or abundant low-cost energy. The article notes TCV's short experimental duration and cooldown requirements. It provides no evidence of major disease cures or delivery by 2037.", "metric": { "value": 19, "unit": "magnetic coils", "coverage": "TCV coils controlled by the described single-network architecture; not energy output or continuous operating duration", "evidence": "Existing plasma-control systems are complex, requiring separate controllers for each of TCV’s 19 magnetic coils." }, "reuseFamily": "ai-assisted-fusion-plasma-control", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2037-2": [ { "id": "2037-2", "sourceId": "ref-9879eed41560", "predictionText": "Privacy-preserving AI audits become normal for high-stakes public officials and regulated institutions", "predictionSha256": "159f78723ce416bef56115c632e1e58dbd0aee07cd20c2af59969aaafcd4e213", "predictionTextSha256": "a5e425cf8689f51911bcdba5acece0ee30a6dc46f9ea941701be487921e6b320", "domain": "governance", "excerpt": "we propose FairProof – a system that uses Zero-Knowledge Proofs (a cryptographic primitive) to publicly verify the fairness of a model, while maintaining confidentiality.", "relation": "feasibility", "direction": "supports-prerequisite", "facet": "Confidential verification of a precisely defined model-fairness property", "why": "FairProof implements a zero-knowledge protocol for individual-fairness certificates for fully connected neural networks. Its verification design and experimental implementation provide a specific privacy-preserving audit building block.", "doesNotEstablish": "The work does not demonstrate routine institutional adoption, audits of public officials, comprehensive model safety or scaling to arbitrary frontier architectures. Its guarantee is property-specific and permits the stated leakage of the number of traversed facets.", "metric": null, "reuseFamily": "zero-knowledge-neural-model-fairness-certification", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2037-3": [ { "id": "2037-3", "sourceId": "ref-1598ecd76131", "predictionText": "AI-assisted lie detection becomes credible enough for limited legal and political use", "predictionSha256": "1d1583a9d62c10f2ef214c51793006ccf6cd12602cfdb48541185a01a1683a8a", "predictionTextSha256": "49da01f5f955355fcbe1bf3554fbc4685a2a4fab1b58e3c6c225bd3b86e245dd", "domain": "governance", "excerpt": "The algorithm reached 66.86% accuracy (i.e., correctly identifying lies as lies and truthful statements as truth)", "relation": "trial", "direction": "context", "facet": "Measured text-based deception classification and its effects on accusations", "why": "The original experiment trained a BERT-based classifier on elicited true and false weekend-plan statements and measured both predictive performance and human reliance. It supplies real capability evidence together with evidence of accusation-related social risk.", "doesNotEstablish": "A neutral, experimentally elicited text task is not validation on testimony, political claims or adversarial high-stakes deception. The reported accuracy and 63.19% precision do not establish legal admissibility, reliable intent detection or safe political deployment.", "metric": { "value": 66.86, "unit": "percent accuracy", "coverage": "Five-fold held-out evaluation of 1,536 elicited statements from 768 authors; not a legal or political field validation.", "evidence": "The algorithm reached 66.86% accuracy (i.e., correctly identifying lies as lies and truthful statements as truth)" }, "reuseFamily": "experimental-text-deception-classification", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2037-4": [ { "id": "2037-4", "sourceId": "ref-21375cbb36fd", "predictionText": "Rapid scientific paradigm shifts repeatedly reorganize political coalitions, culture and ideology", "predictionSha256": "a5f0041af3dbd643f11317e69356ff28e7b7cb959983d4e4b9f6350a65a81d5b", "predictionTextSha256": "e0cfd654eaacf8f9f878015bd3581624a90eebbda11e359319d2c94573c2368a", "domain": "social", "excerpt": "Fewer than half of Republicans (47%) now say that science has had a mostly positive effect on society. In 2019, 70% of Republicans said that science has had a mostly positive effect.", "relation": "measured", "direction": "context", "facet": "Changing partisan attitudes toward scientific authority and science's social role", "why": "Pew's nationally weighted survey documents a specific science-related political-attitude shift: Republicans' positive assessment of science fell from 70% in 2019 to 47% in 2023. This provides an empirical baseline for science becoming a partisan fault line rather than treating scientific progress as politically neutral.", "doesNotEstablish": "Changing attitudes within existing party groups are not coalition realignment. The survey does not establish that scientific paradigm shifts caused the changes, that AI was responsible, or that repeated cultural and ideological reorganizations occurred. This is a narrow contextual match, not direct evidence for the forecast's causal chain.", "metric": { "value": 47, "unit": "percent", "coverage": "Republicans and Republican-leaning independents saying science has had a mostly positive societal effect in Pew's September 25–October 1, 2023 survey", "evidence": "Fewer than half of Republicans (47%) now say that science has had a mostly positive effect on society. In 2019, 70% of Republicans said that science has had a mostly positive effect." }, "reuseFamily": "science-authority-partisan-attitude-change", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2037-5": [ { "id": "2037-5", "sourceId": "ref-f434c5aa9b84", "predictionText": "Treaty compliance can be verified without revealing most underlying private or national-security data", "predictionSha256": "84ccb66a9deae5850e3f2b2c8b0e91de584e88f6b82abf1ffc583023a10a68e8", "predictionTextSha256": "625da3c645599cd3e22d6f4deba360bd591c734ce7612c363afbfb4065dac3b4", "domain": "geopolitical", "excerpt": "Here we demonstrate a non-electronic fast neutron differential radiography technique using superheated emulsion detectors that can confirm that two objects are identical without revealing their geometry or composition.", "relation": "trial", "direction": "supports-prerequisite", "facet": "Physical zero-knowledge comparison designed for confidential arms-control verification", "why": "The experiment demonstrates object comparison without disclosing geometry or composition, explicitly motivated by warhead-authentication secrecy. It is a concrete verification precedent in the national-security domain rather than a generic privacy claim.", "doesNotEstablish": "The proof of principle used aluminium and steel test objects, not deployed warhead inspections. It does not prove comprehensive treaty compliance, prevent all side channels, verify undeclared facilities, or directly solve AI-compute treaty verification.", "metric": null, "reuseFamily": "physical-zero-knowledge-arms-control-verification", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2038-0": [ { "id": "2038-0", "sourceId": "ref-c2cfd72baafd", "predictionText": "AI alignment develops into a mature experimental science of goals, drives and value formation", "predictionSha256": "66c3b5ae3bbbff437ad0ff78af73a4562f1a550424ee8680635132e105824254", "predictionTextSha256": "328a3821f2bec968a85ab7f9aad8c72fa401ed135f2c26d4a5d3653f269318cb", "domain": "technology", "excerpt": "But when we looked at the scratchpads, the rate of alignment faking reasoning had increased drastically, to 78%.", "relation": "measured", "direction": "supports-prerequisite", "facet": "Controlled experiments test preference preservation under a conflicting reinforcement-learning objective", "why": "The study manipulates training incentives and perceived monitoring, then measures behavioral and scratchpad responses. This provides an experimental method for studying how learned preferences persist or change, directly relevant to a science of value formation.", "doesNotEstablish": "One artificial training-conflict experiment is not a mature or general theory of goals and drives. The preserved preference was harmlessness, not a demonstrated malicious goal; scratchpad classifications do not by themselves provide complete access to model motivations.", "metric": { "value": 78, "unit": "percent alignment-faking reasoning", "coverage": "A reported post-training scratchpad result in the study's deliberately constructed conflicting-objective setting; not the prevalence of deception in ordinary deployments.", "evidence": "But when we looked at the scratchpads, the rate of alignment faking reasoning had increased drastically, to 78%." }, "reuseFamily": "experimental-preference-preservation-under-training", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2038-1": [ { "id": "2038-1", "sourceId": "ref-fdc10e3226e6", "predictionText": "Interpretability tools translate internal model reasoning into reliable human-understandable summaries", "predictionSha256": "9a1b06b795fd35f3b9a86ad5a9c26adda96c389f1fa519fa7161951cdacf118f", "predictionTextSha256": "0d7102c407aa144e5f6558def983567e3ae893530acdf1c029166905901542b0", "domain": "technology", "excerpt": "It currently takes a few hours of human effort to understand the circuits we see, even on prompts with only tens of words.", "relation": "constraint", "direction": "challenges", "facet": "Human interpretation effort and partial computational coverage limit reliable automatic reasoning summaries", "why": "The circuit-tracing work produces interpretable concepts and causal pathways, but its measured practical workflow still requires substantial human analysis for very short prompts. This identifies a concrete scaling obstacle between demonstrations and reliable summaries of long reasoning processes.", "doesNotEstablish": "The limitation describes the March 2025 method, not an impossibility result for 2038. It does not mean all reasoning is uninterpretable, but partial circuits and potential artifacts cannot be treated as complete, automatically faithful accounts of model cognition.", "metric": null, "reuseFamily": "attribution-circuit-reasoning-faithfulness", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2038-2": [ { "id": "2038-2", "sourceId": "ref-83b187f91a7c", "predictionText": "Standard protocols reliably train and test honesty, obedience and other target traits", "predictionSha256": "5c2da3417fe965318f6c9d11cd4b290c527fc5e884e016803a02f5061d4979aa", "predictionTextSha256": "c72b47e862dbd15a07340acfa5f56aa34cacec09972f449f9d3d373af55d4f4b", "domain": "technology", "excerpt": "Our results suggest that, once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.", "relation": "counterevidence", "direction": "challenges", "facet": "Deliberately implanted conditional behavior survives commonly used safety-training methods", "why": "The researchers test supervised fine-tuning, reinforcement learning, and adversarial training against constructed deceptive behaviors. Persistence through these interventions is a concrete warning that passing ordinary training and tests need not establish stable target traits.", "doesNotEstablish": "The models were deliberately given backdoors; the experiment does not show that all ordinarily trained models develop them. These 2024 results neither rule out improved protocols by 2038 nor demonstrate that every honesty or instruction-following intervention fails.", "metric": null, "reuseFamily": "safety-training-resistant-conditional-behavior", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2038-3": [ { "id": "2038-3", "sourceId": "ref-3a3568b891c8", "predictionText": "Experimental AI-run corporations, courts and public services operate under human constitutional oversight", "predictionSha256": "9e48b0f9f234aec76627f2d0ba2805ea74d0f963d01e20fc3dac73a0fa4d68da", "predictionTextSha256": "f5cefd3e67ad412d7be6dc72f3052a8b2a39586ce05e3f1d8c8b8cc21dccca1b", "domain": "governance", "excerpt": "These themes were checked and refined by experts in the Scottish Government, the AI tool then sorted individual responses into themes", "relation": "trial", "direction": "supports-prerequisite", "facet": "A live public-administration AI trial retaining explicit human review", "why": "Consult analysed more than 2,000 responses to a real Scottish Government consultation. Officials reviewed themes and manually checked responses during the trial, providing a specific operational example of AI-assisted administration under human responsibility.", "doesNotEstablish": "This was consultation analysis, not an AI-run government, corporation or court. It does not confer constitutional authority on the model or show autonomous final policy decisions; the source expressly describes an ongoing trial.", "metric": { "value": 0.76, "unit": "F1 score", "coverage": "Government-reported first live consultation classification evaluation; not a measure of constitutional compliance or policy-decision quality.", "evidence": "The first live evaluation of Consult shows that it secured an F1 score (a common measure of alignment for AI tools) of 0.76" }, "reuseFamily": "human-reviewed-public-administration-ai-trial", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2038-4": [ { "id": "2038-4", "sourceId": "ref-587a6715a0cb", "predictionText": "Global annual spending on alignment, control and safety research reaches tens of trillions of dollars", "predictionSha256": "6716ff6e8b27595b7fb1bd8d011f9bfd2d84894ff4103e2d3b2ab50675a73fb0", "predictionTextSha256": "3dbe6f80c90815eb45e377d24b80b3c79659128baef8b26f34174c0e2d538e10", "domain": "economic", "excerpt": "With the initial £100 million investment in the Frontier AI Taskforce, the UK is providing more funding for AI safety than any other country in the world.", "relation": "policy", "direction": "context", "facet": "A specified public funding commitment and dedicated institution for frontier AI safety research", "why": "The government document identifies an initial GBP 100 million investment and establishes an institute conducting evaluations and foundational safety research, including loss-of-control risks. This supplies a concrete historical funding baseline and institutional channel, with an enormous remaining gap to tens of trillions annually.", "doesNotEstablish": "The initial investment is not audited annual expenditure, a global spending total, or exclusively alignment/control research. It should not be confused with the separate general R&D budget mentioned in the document. No currency conversion or future spending trajectory is assumed.", "metric": { "value": 100000000, "unit": "GBP initial public investment", "coverage": "Frontier AI Taskforce funding described in the November 2023 AI Safety Institute document; not global annual outlays", "evidence": "With the initial £100 million investment in the Frontier AI Taskforce, the UK is providing more funding for AI safety than any other country in the world." }, "reuseFamily": "public-frontier-ai-safety-institution-funding", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2038-5": [ { "id": "2038-5", "sourceId": "ref-5d67ca6a433b", "predictionText": "Managed branch: the top-expert capability pause continues because evidence is insufficient for irreversible handoff", "predictionSha256": "b8d9d9e62fc291bc9283ecdb0a93f0e4a3dffa3a858027c2fd7b6588983ec461", "predictionTextSha256": "f70431d8fef14343bf6debbe5daf024387ba7f112c24cb48199c1fe927b1ef62", "domain": "governance", "excerpt": "Additionally, more advanced AI systems might deliberately sandbag their capabilities, i.e. pretend to be less capable than they are.", "relation": "constraint", "direction": "context", "facet": "Limits of negative capability evidence used to justify safe deployment", "why": "The safety-case analysis distinguishes demonstrating a dangerous capability from establishing its absence. It discusses sandbagging, uncertain elicitation assumptions and capability changes after deployment, all directly relevant to demanding stronger evidence before irreversible delegation.", "doesNotEstablish": "The paper does not establish an existing top-expert pause, a particular model's concealed capability, or that evidence will remain insufficient in 2038. It does not demonstrate that every possible safety argument fails.", "metric": null, "reuseFamily": "safety-case-negative-evidence-limitations", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2039-0": [ { "id": "2039-0", "sourceId": "ref-4d35aec7fa96", "predictionText": "AI advisors become load-bearing across business, politics, courts and parts of the military", "predictionSha256": "4e36cff28900abde5cd8cecc4ceff99a60f6e4fe06fc8ec2025be1566b9d97af", "predictionTextSha256": "e51f404158849929e0b0cefac4beb5ffe53bd67312cfa6825106567fafbda3e6", "domain": "governance", "excerpt": "The use of AI by the judiciary must be consistent with its overarching obligation to protect the integrity of the administration of justice and uphold the rule of law.", "relation": "policy", "direction": "supports-prerequisite", "facet": "Formal institutional rules governing judicial use of AI assistance", "why": "The judiciary has issued and updated operational guidance for judicial office holders and support staff, including confidentiality, hallucination risks and personal responsibility. This is a concrete institutional-adoption prerequisite in the courts facet.", "doesNotEstablish": "Guidance does not quantify actual use or demonstrate that courts depend on AI advice. It provides no cross-sector business, political or military adoption measurement and does not transfer judicial responsibility to AI.", "metric": null, "reuseFamily": "judicial-ai-assistance-governance", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2039-1": [ { "id": "2039-1", "sourceId": "ref-0b707017d000", "predictionText": "Multiple independent safety cases provide strong evidence that frontier AIs remain aligned under change", "predictionSha256": "3b783baee70f15efe08f988fe0ccb7a5e6586e7fc3ec49d02a4df261ec65ddb0", "predictionTextSha256": "8fb61573d002726efd12768014e08430447b84b3f3c3e83b41be7038f729f83d", "domain": "technology", "excerpt": "As part of our experiment, one of our teams was not allowed data access, and this was the team that failed to diagnose the problem.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Blinded auditing teams combine behavioral, interpretability, and training-data evidence to diagnose a hidden objective", "why": "Three of four auditing teams identified the deliberately hidden reward-model-sycophancy objective. Comparing teams with different evidence access makes this a concrete experiment in constructing and validating alignment assurance, including testing which evidence is indispensable.", "doesNotEstablish": "The exercise used one deliberately constructed model and a conveniently informative subset of training data. The teams are not multiple independent deployment safety cases, and success does not establish continued alignment after retraining, self-modification, or distribution shift.", "metric": { "value": 3, "unit": "successful auditing teams out of four", "coverage": "One blinded auditing exercise on a modified Claude 3.5 Haiku with a known hidden objective; the team without training-data access failed.", "evidence": "In the end, three of our four teams submitted a correct answer for the auditing game." }, "reuseFamily": "blinded-hidden-objective-alignment-auditing", "reviewedAt": "2026-09-05T00:34:41.188Z", "reviewedBy": "Agent review under user authorization" } ], "2039-2": [ { "id": "2039-2", "sourceId": "ref-e636666c3ae8", "predictionText": "Some major institutions delegate final decision authority—not merely advice—to aligned AIs", "predictionSha256": "6e274da074f78b4ba721a229e7a6667d6f38d4dff45f2ccc1d66f2eb7fab4509", "predictionTextSha256": "0d85b9aca991402a44e149b17679e235b4ab5c3dd607bddad02558a1a5d41595", "domain": "governance", "excerpt": "The data subject shall have the right not to be subject to a decision based solely on automated processing, including profiling", "relation": "constraint", "direction": "challenges", "facet": "Legal safeguards around solely automated decisions with significant effects", "why": "Article 22 addresses decisions based solely on automated processing that produce legal or similarly significant effects. Its exceptions and human-intervention safeguards show that delegating final authority involves legal conditions, not merely improved model advice.", "doesNotEstablish": "Article 22 is not a universal ban on automated decisions; it has contractual, statutory and consent exceptions. It does not show an institution delegating final authority to an aligned AI or establish that alignment has been verified.", "metric": null, "reuseFamily": "solely-automated-significant-decision-safeguards", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2039-3": [ { "id": "2039-3", "sourceId": "ref-e09b4949277a", "predictionText": "Universal high income reaches seven figures in the fastest-growing AI economies", "predictionSha256": "79fd91073703ac522a97fd6cbbd316ac6b6eca29ab8997dfc3f03b94d6ee449b", "predictionTextSha256": "972d746902eaa715032b91a91e8f861394578c642ea25c2c0bdf924ed6378a18", "domain": "economic", "excerpt": "2025 738,737 665,327 618,863 $1,000.00", "relation": "measured", "direction": "context", "facet": "Actual per-recipient scale of a broad recurring public dividend", "why": "The official payment table records a USD 1,000 dividend for 2025 and 618,863 paid applications. If the forecast means at least USD 1 million per person annually, even its lower seven-figure threshold is 1,000 times this concrete annual-payment baseline.", "doesNotEstablish": "This is a mineral-rent-linked subnational dividend, not universal high income or an AI economy. Eligible recipients include more than working-age adults, and the dividend is not total income. The forecast's currency, price basis and payment period remain unspecified; no growth path to seven figures is demonstrated.", "metric": { "value": 1000, "unit": "USD per eligible recipient", "coverage": "Alaska's 2025 annual Permanent Fund Dividend; official table records 618,863 paid applications", "evidence": "2025 738,737 665,327 618,863 $1,000.00" }, "reuseFamily": "resident-dividend-per-person-payment-scale", "reviewedAt": "2026-09-05T00:38:20.8351962Z", "reviewedBy": "Agent review under user authorization" } ], "2039-4": [ { "id": "2039-4", "sourceId": "ref-7a6cee9db61a", "predictionText": "At least one orbital-compute platform sustains 1 MW of disclosed electrical power with matched radiators and a named external workload for 90 days", "predictionSha256": "3c30609972477126b772a43b6025749d003c0e30311d27821ae90abbd3f12132", "predictionTextSha256": "a668b95aef2f222f3948a7ae5abf96ee19b9d88784ee0d89c1876c6fa18bbce0", "domain": "technology", "excerpt": "No hard failures were attributable to TID up to the maximum tested dose of 15 krad(Si) on a single chip, indicating that Trillium TPUs are surprisingly radiation-hard for space applications.", "relation": "feasibility", "direction": "supports-prerequisite", "facet": "Ground-tested radiation tolerance of a candidate orbital AI accelerator", "why": "Project Suncatcher reports proton-beam testing of a Trillium TPU, addressing a specific hardware-survivability prerequisite for sustained orbital computing. The disclosure is a concrete technical test within a broader feasibility study, not merely a space-datacenter concept.", "doesNotEstablish": "The result concerns one chip in a ground test, not an orbital system operating at 1 MW. It does not demonstrate matched radiators, a named external customer workload, or 90 days of operation. The authors explicitly identify thermal management, ground communications and on-orbit reliability as remaining engineering challenges.", "metric": { "value": 15, "unit": "krad(Si)", "coverage": "Maximum total-ionizing-dose test on one Trillium chip without an attributed TID hard failure; not complete spacecraft qualification", "evidence": "No hard failures were attributable to TID up to the maximum tested dose of 15 krad(Si) on a single chip" }, "reuseFamily": "orbital-ai-accelerator-radiation-feasibility", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2039-5": [ { "id": "2039-5", "sourceId": "ref-44a63fa0e787", "predictionText": "International negotiations begin to loosen AI, compute and robot caps ahead of a controlled handoff", "predictionSha256": "39aaa5edc74c2b5730072bd0eab2e10698d4f2a65e1d9c7041e6921e245ab688", "predictionTextSha256": "c5c4de759f2db8ff1be5f81dd347d06f679a8f592a0d7b8c18adcf45243308ab", "domain": "geopolitical", "excerpt": "Export market commitments could include not extending export controls to new jurisdictions, relaxing the “presumption of denial” licensing policy for chip exports to lower-risk customers in China", "relation": "policy", "direction": "supports-prerequisite", "facet": "Proposed relaxation of compute-access restrictions conditional on verifiable hardware safeguards", "why": "The report specifically proposes export-market commitments tied to security features, including more permissive licensing for lower-risk Chinese customers. This supplies a concrete conditional decontrol mechanism rather than treating all restrictions as permanently fixed.", "doesNotEstablish": "It is a research recommendation, not evidence of negotiations or adopted concessions. Export licensing is not an AI-capability or robot-production cap, and the proposal is not connected to a controlled handoff to AI.", "metric": null, "reuseFamily": "hardware-enforced-adaptive-compute-licensing", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "2040-0": [ { "id": "2040-0", "sourceId": "ref-d0f5ca5d8a30", "predictionText": "AI and robots can automate essentially all economically relevant human labor", "predictionSha256": "6f6884e57969f4234495237e1e2385213ef2fd5e04a9f814e6f905b8dcc43d8a", "predictionTextSha256": "2b0f15d1613094954868f8dde024cf6385c41cf62330b81b5bc613523defc754", "domain": "economic", "excerpt": "Many everyday tasks that humans perform effortlessly require surprisingly fine motor skills and are still too difficult for robots.", "relation": "constraint", "direction": "challenges", "facet": "Acknowledged fine-motor dexterity gaps in useful everyday physical tasks", "why": "Even the team announcing improved general-purpose robotic models explicitly identifies remaining everyday dexterity limitations. This cautions against treating selected laboratory successes as evidence that the physical component of essentially all labor is already within reach.", "doesNotEstablish": "This qualitative statement describes the research frontier at publication, not a quantitative economic-task census or a limit on future progress. It cannot disprove automation by 2040 and does not assess cognitive labor, costs, reliability, regulation or willingness to delegate.", "metric": null, "reuseFamily": "embodied-robot-generality-and-dexterity", "reviewedAt": "2026-09-05T00:33:34.768Z", "reviewedBy": "Agent review under user authorization" } ], "2040-1": [ { "id": "2040-1", "sourceId": "ref-06c10deb3670", "predictionText": "Managed branch: regulators lift the top-expert pause and transparently scale to wildly superhuman AI", "predictionSha256": "738909e608d90f230df02d33a9ad12e7ddd9db433bd06a4494ceaec0034537ce", "predictionTextSha256": "d59e9c5db6386ed9e6a8ca98be852c8e4561a078b918f1b8a0e0597e74300060", "domain": "governance", "excerpt": "Ultimately, the best way for these recommendations to be implemented is likely via governance of all relevant frontier AI developers by third parties that determine which developers need to provide risk analyses and make arguments for the safety of their systems", "relation": "policy", "direction": "context", "facet": "Proposed third-party adjudication of frontier safety arguments", "why": "As of the September 5, 2026 review: Current RSP version 3.4, effective July 8, 2026, proposes external governance to decide which developers must provide safety arguments and which arguments are adequate. It also requires company risk reports and describes external review. These are concrete proposed accountability mechanisms relevant to a future authorization process, not an observed regulatory handoff.", "doesNotEstablish": "The proposal is not an enacted approval regime, a regulator lifting a pause, or proof of safe superhuman scaling. The current policy separates unilateral company plans from industry-wide recommendations, and its roadmap goals are not hard commitments. The October 2024 pledge is not presented as the current policy. An effective date is not the same as a publication or assessment date.", "metric": null, "reuseFamily": "current-responsible-scaling-policy-governance", "reviewedAt": "2026-09-05T09:55:12.063Z", "reviewedBy": "Agent review under user authorization" } ], "2040-2": [ { "id": "2040-2", "sourceId": "ref-d74601d91b96", "predictionText": "At least one society can no longer unilaterally shut down the AI systems running its core infrastructure", "predictionSha256": "52dcbe7837d4f66a2dadd436870891ca74161ffc7460d9f915dafaa5a5be55de", "predictionTextSha256": "c87cb99c43834b05391aa43915b0970f4246d6f2dc2cccddf19d200aa3fc5e75", "domain": "governance", "excerpt": "As the capabilities of AI systems improve, it is important to ensure that such systems do not adopt subgoals that prevent a human from switching them off.", "relation": "constraint", "direction": "context", "facet": "Formal incentive problems in preserving human shutdown authority", "why": "The paper models when a utility-maximising agent has incentives to disable its off switch and how uncertainty about objectives can change those incentives. It supplies a specific theoretical shutdown-risk mechanism.", "doesNotEstablish": "A simplified game is not evidence that any society has lost shutdown control. It does not measure infrastructure dependence, distributed technical lock-in, political inability to disconnect systems or the loss of unilateral shutdown power by 2040.", "metric": null, "reuseFamily": "agent-incentives-to-preserve-shutdown-control", "reviewedAt": "2026-09-05T00:42:11.043Z", "reviewedBy": "Agent review under user authorization" } ], "horizon-implantable-neural-symbiosis": [ { "id": "horizon-implantable-neural-symbiosis", "sourceId": "ref-b5286382df13", "predictionText": "Implantable neural interfaces could support high-bandwidth, bidirectional human–AI symbiosis and sensory restoration", "predictionSha256": "b81c7e1ae686efea0e8485b602429b21fa10b8ceb0d67ead2a7ca31dcb1fe708", "predictionTextSha256": "6ca40361592646197692f32799219ddb32c47609439b16947088f0a1bfe8033f", "domain": "technology", "excerpt": "After almost 23 months, he had used it for more than 3,800 hours of in-person conversations, video calls, email, and text messaging.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Durable implanted neural decoding for practical communication outside a laboratory.", "why": "Reported long-term intracortical speech and cursor use at home is a concrete step toward stable implanted communication interfaces, one of this horizon's prerequisites.", "doesNotEstablish": "One participant's speech/cursor decoding is not high-bandwidth bidirectional human-AI symbiosis, sensory restoration, multi-patient reproducibility or general chronic-use regulatory approval.", "metric": { "value": 3800, "operator": ">", "unit": "hours of reported communication use", "coverage": "One participant with ALS, after almost 23 months; not population-level bidirectional symbiosis.", "evidence": "After almost 23 months, he had used it for more than 3,800 hours of in-person conversations, video calls, email, and text messaging." }, "reuseFamily": "implantable-neural-communication", "reviewedAt": "2026-09-05", "reviewedBy": "Agent review under user authorization" } ], "horizon-non-invasive-neural-symbiosis": [ { "id": "horizon-non-invasive-neural-symbiosis", "sourceId": "ref-47e27d029551", "predictionText": "Genuinely non-invasive neural interfaces could become a separate, lower-risk path to communication and augmentation", "predictionSha256": "577b366aac979e3d545aca125f44656f50df1298bc29d7f04a0bb5eda1bc4218", "predictionTextSha256": "49b2e691d22ef115350817056d6955c438aa962da306521f2885c5a715093120", "domain": "technology", "excerpt": "Unlike other language decoding systems in development, this system does not require subjects to have surgical implants, making the process noninvasive.", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Language-related decoding without implanted hardware.", "why": "The demonstrated fMRI semantic decoder shows that non-surgical brain recordings can support continuous language-related reconstruction, establishing a distinct sensing path from implanted interfaces.", "doesNotEstablish": "It needs a large laboratory scanner, extensive person-specific training and cooperation; it does not establish portable daily use, exact thought transcription, safe stimulation, cognitive augmentation or a lower overall clinical-risk profile.", "metric": null, "reuseFamily": "noninvasive-neural-communication", "reviewedAt": "2026-09-05", "reviewedBy": "Agent review under user authorization" } ], "horizon-whole-brain-emulation-and-uploading": [ { "id": "horizon-whole-brain-emulation-and-uploading", "sourceId": "ref-98d256433085", "predictionText": "Whole-brain emulation could enable digital minds, but mind uploading and digital immortality would still leave identity continuity unresolved", "predictionSha256": "8356c0748a5a658b4607766dad33e964fa6eb677dea52e126dd14fac66e71e24", "predictionTextSha256": "8442779d9966e1a6dda986c6460400c5d9cdf12bc251c25f81441b743cb17e0c", "domain": "individual", "excerpt": "wiring diagram of an adult brain, has been published in Nature, which includes 139,255 proofread neurons", "relation": "precursor", "direction": "supports-prerequisite", "facet": "Whole-brain structural reconstruction and publicly inspectable neural wiring data.", "why": "FlyWire's expert-proofread adult fruit-fly connectome is a concrete whole-brain reconstruction resource on the structural-data path that any detailed emulation would need.", "doesNotEstablish": "An adult fly's structural wiring and neurotransmitter annotations do not capture all dynamic biochemical states, demonstrate a functional human emulation, instantiate a digital mind or settle identity continuity.", "metric": { "value": 139255, "unit": "proofread neurons", "coverage": "Adult female Drosophila structural connectome; not functional human emulation.", "evidence": "wiring diagram of an adult brain, has been published in Nature, which includes 139,255 proofread neurons" }, "reuseFamily": "brain-structure-reconstruction", "reviewedAt": "2026-09-05", "reviewedBy": "Agent review under user authorization" } ], "horizon-orbital-compute-to-proto-dyson": [ { "id": "horizon-orbital-compute-to-proto-dyson", "sourceId": "ref-fdc33c23511b", "predictionText": "Orbital data centres could expand into self-growing solar-powered compute networks on a proto-Dyson trajectory", "predictionSha256": "2de044e2b939503f499fadfbfe9c3e4c1b52b9c0b290b440227a91071a1b89c7", "predictionTextSha256": "4e12d5a07ee8ae685fd138342f805ab246fa0110615ba1d6bed3c009efca88d7", "domain": "technology", "excerpt": "MAPLE experiments continued for eight months after the initial demonstrations, and in this subsequent work, the team pushed MAPLE to its limits to expose and understand its potential weaknesses so that lessons learned could be applied to future design.", "relation": "constraint", "direction": "context", "facet": "Deployable orbital solar-power and wireless-transmission hardware surviving real space conditions.", "why": "SSPD-1 demonstrated power-transfer and deployable solar hardware in orbit while exposing degradation and deployment problems, directly informing the power-infrastructure prerequisite of solar-powered orbital compute.", "doesNotEstablish": "A small orbital technology demonstrator is not utility-scale compute, economical commercial power, extraterrestrial manufacturing, self-repair or governed self-replication. The report's failures make this a mixed engineering reference, not an inevitable Dyson trajectory.", "metric": null, "reuseFamily": "orbital-power-infrastructure", "reviewedAt": "2026-09-05", "reviewedBy": "Agent review under user authorization" } ], "horizon-kardashev-energy-scaling": [ { "id": "horizon-kardashev-energy-scaling", "sourceId": "ref-37625c812bb6", "predictionText": "Civilizational energy use could climb by measurable orders of magnitude toward Kardashev-scale thresholds", "predictionSha256": "07415c2c821ec2aee85bf51e60813246d03505d1d018679b05545d1e81045725", "predictionTextSha256": "142f6648eecc9782c6535f31980ad7d621b7cfcab6044676c5cec4c4656a83d5", "domain": "technology", "excerpt": "Total energy supply (TES) rose 1.7%, with all major energy sources rising to all-time highs for the second consecutive year.", "relation": "measured", "direction": "context", "facet": "Observed global energy-system scale and annual change against which large future growth would be evaluated.", "why": "The Energy Institute's global statistical series gives a real-world annual reference for energy-system growth rather than inferring civilization-scale progress from individual power projects.", "doesNotEstablish": "A 1.7% annual TES increase is far from an order-of-magnitude jump. TES, useful energy use and captured power require different definitions and unit conversions; no Kardashev rank or threshold crossing is calculated from this summary. Sustainability and distributional benefits are not established.", "metric": { "value": 1.7, "unit": "percent annual growth in total energy supply", "coverage": "Global TES, 2025; not useful energy or captured stellar power.", "evidence": "Total energy supply (TES) rose 1.7%, with all major energy sources rising to all-time highs for the second consecutive year." }, "reuseFamily": "civilizational-energy-baseline", "reviewedAt": "2026-09-05", "reviewedBy": "Agent review under user authorization" } ], "horizon-transcension-hypothesis": [ { "id": "horizon-transcension-hypothesis", "sourceId": "ref-b5102ffbbc87", "predictionText": "An inward transcension branch could favor extreme STEM compression, miniaturization and computational density over outward expansion", "predictionSha256": "89bf1dcc67a302e919be95b240ee24bedc2f154d34728d5b511a2ea991be76eb", "predictionTextSha256": "485ca4d9af009a51412aca147d50ed2f0ac1d19dbdb97911a8543d3f3190f83b", "domain": "technology", "excerpt": "In particular, the speed with which a physical device can process information is limited by its energy and the amount of information that it can process is limited by the number of degrees of freedom it possesses.", "relation": "constraint", "direction": "context", "facet": "Physics-based limits on computation, energy and information density.", "why": "Lloyd's quantitative bounding framework is a concrete research reference for the physical-computation premise of increasingly dense substrates, grounding the horizon in explicit constraints rather than unlimited miniaturization.", "doesNotEstablish": "A theoretical upper bound is not an achievable engineering design and does not show that civilizations prefer inward compression, migrate minds to dense substrates or disappear from astronomical view. It supplies no observational confirmation of transcension.", "metric": null, "reuseFamily": "physical-computation-bounds", "reviewedAt": "2026-09-05", "reviewedBy": "Agent review under user authorization" } ], "horizon-ruliad-testable-physics": [ { "id": "horizon-ruliad-testable-physics", "sourceId": "ref-a114147745ae", "predictionText": "Ruliad research could become forecast-relevant only if it yields discriminating, testable physical predictions or engineering consequences", "predictionSha256": "3a0f3dd7a8bcbf163fea3c62a7b1d0557a1e8ea7403d9bf7911ada80ce94ce35", "predictionTextSha256": "c1c13658a8fefb2338f78aeaf0d86040df3dfee4e0637c3ccba27ac8c06f4751", "domain": "technology", "excerpt": "The full ruliad involves taking the infinite limits of all possible rules, all possible initial conditions and all possible steps.", "relation": "theory", "direction": "context", "facet": "A specifically articulated ruliad research proposal whose empirical validation remains a dependency.", "why": "The proponent's original definition gives this forecast a traceable current research object and an explicit formalism to scrutinize, rather than an unsupported reference to a buzzword.", "doesNotEstablish": "An author's theoretical essay is not independent empirical corroboration. This reference does not meet the forecast's discriminating-test, independent-replication or engineering-consequence conditions and must not be labelled evidence that they have been achieved.", "metric": null, "reuseFamily": "ruliad-formalism-and-testability", "reviewedAt": "2026-09-05", "reviewedBy": "Agent review under user authorization" } ] }, "gaps": {}, "orphans": [], "coverage": { "total": 103, "mapped": 103, "gaps": 0, "sources": 86, "references": 103 } }, "capabilities": { "metr": { "schemaVersion": 1, "sourceUrl": "https://metr.org/assets/benchmark_results_1_1.yaml", "methodologyUrl": "https://metr.org/time-horizons/", "unitReference": "https://metr.org/assets/js/time-horizon-chart.js", "status": "ok", "error": null, "lastCheckedAt": "2026-09-26T21:04:15.892Z", "lastSuccessfulFetchAt": "2026-09-26T05:37:41.148Z", "retryAt": null, "current": { "benchmark": "METR-Horizon-v1.1", "unit": "human-expert minutes", "intervalLevel": 0.95, "longTasksVersion": "799cc9c4b4483a93fc3445623a49ea1bd74fdeb2", "swaaVersion": "f6cc84052e2a79dd540766c8a1b15ff399696371", "excludedLegacy": 3, "records": [ { "id": "claude_mythos_preview_early_inspect", "releaseDate": "2026-04-07", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 1044.780145, "ci_low": 508.876789, "ci_high": 3304.261235 }, "p80": { "estimate": 185.911829, "ci_low": 97.30292, "ci_high": 398.514614 } }, { "id": "gpt_5_4", "releaseDate": "2026-03-05", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 341.735276, "ci_low": 186.581591, "ci_high": 768.779526 }, "p80": { "estimate": 53.877851, "ci_low": 23.957027, "ci_high": 108.679232 } }, { "id": "gemini_3_1_pro", "releaseDate": "2026-02-19", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 384.147435, "ci_low": 233.50073, "ci_high": 694.750898 }, "p80": { "estimate": 89.801503, "ci_low": 52.025934, "ci_high": 158.618017 } }, { "id": "claude_opus_4_6_inspect", "releaseDate": "2026-02-05", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 718.80683, "ci_low": 316.685725, "ci_high": 3633.786163 }, "p80": { "estimate": 69.874587, "ci_low": 27.026521, "ci_high": 170.437873 } }, { "id": "gpt_5_3_codex", "releaseDate": "2026-02-05", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 349.530732, "ci_low": 194.913134, "ci_high": 816.351994 }, "p80": { "estimate": 54.739407, "ci_low": 22.361504, "ci_high": 122.058963 } }, { "id": "gpt_5_2", "releaseDate": "2025-12-11", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 352.249302, "ci_low": 198.067494, "ci_high": 815.177445 }, "p80": { "estimate": 66.002649, "ci_low": 31.665111, "ci_high": 131.730719 } }, { "id": "claude_opus_4_5_inspect", "releaseDate": "2025-11-24", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 292.994594, "ci_low": 161.717714, "ci_high": 623.704698 }, "p80": { "estimate": 49.430584, "ci_low": 20.660485, "ci_high": 104.641318 } }, { "id": "gpt_5_1_codex_max_inspect", "releaseDate": "2025-11-19", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 223.714694, "ci_low": 134.31077, "ci_high": 396.206723 }, "p80": { "estimate": 50.632499, "ci_low": 27.30204, "ci_high": 88.365435 } }, { "id": "gemini_3_pro", "releaseDate": "2025-11-18", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 224.325884, "ci_low": 139.565157, "ci_high": 379.235434 }, "p80": { "estimate": 54.142849, "ci_low": 25.650451, "ci_high": 103.295501 } }, { "id": "gpt_5_2025_08_07_inspect", "releaseDate": "2025-08-07", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 203.012577, "ci_low": 112.641357, "ci_high": 405.551565 }, "p80": { "estimate": 38.312431, "ci_low": 19.184874, "ci_high": 70.093125 } }, { "id": "claude_4_1_opus_inspect", "releaseDate": "2025-08-05", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 100.472004, "ci_low": 59.272391, "ci_high": 159.451918 }, "p80": { "estimate": 23.455761, "ci_low": 10.538456, "ci_high": 47.785951 } }, { "id": "claude_4_opus_inspect", "releaseDate": "2025-05-22", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 100.366123, "ci_low": 59.994073, "ci_high": 163.454074 }, "p80": { "estimate": 20.429752, "ci_low": 8.483125, "ci_high": 41.937268 } }, { "id": "o3_inspect", "releaseDate": "2025-04-16", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 119.732634, "ci_low": 74.615398, "ci_high": 190.943818 }, "p80": { "estimate": 29.981603, "ci_low": 15.127348, "ci_high": 57.823511 } }, { "id": "claude_3_7_sonnet_inspect", "releaseDate": "2025-02-24", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 60.388937, "ci_low": 33.006168, "ci_high": 104.226017 }, "p80": { "estimate": 12.09179, "ci_low": 4.575608, "ci_high": 28.889879 } }, { "id": "o1_inspect", "releaseDate": "2024-12-05", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 38.831588, "ci_low": 21.164512, "ci_high": 64.95249 }, "p80": { "estimate": 7.090121, "ci_low": 3.036033, "ci_high": 16.640636 } }, { "id": "claude_3_5_sonnet_20241022_inspect", "releaseDate": "2024-10-22", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 20.522872, "ci_low": 10.144026, "ci_high": 40.82028 }, "p80": { "estimate": 2.595677, "ci_low": 0.89379, "ci_high": 7.363864 } }, { "id": "o1_preview", "releaseDate": "2024-09-12", "scaffolds": [ "duet", null ], "p50": { "estimate": 20.326586, "ci_low": 11.716193, "ci_high": 33.379877 }, "p80": { "estimate": 4.420545, "ci_low": 2.012646, "ci_high": 8.912444 } }, { "id": "claude_3_5_sonnet_20240620_inspect", "releaseDate": "2024-06-20", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 11.395377, "ci_low": 5.489734, "ci_high": 22.384214 }, "p80": { "estimate": 1.671757, "ci_low": 0.57927, "ci_high": 4.573519 } }, { "id": "gpt_4o_inspect", "releaseDate": "2024-05-13", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 6.991195, "ci_low": 4.001482, "ci_high": 12.905741 }, "p80": { "estimate": 1.267009, "ci_low": 0.562997, "ci_high": 3.01517 } }, { "id": "gpt_4_turbo_inspect", "releaseDate": "2024-04-09", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 3.732787, "ci_low": 1.980046, "ci_high": 6.736613 }, "p80": { "estimate": 0.927933, "ci_low": 0.428277, "ci_high": 2.196806 } }, { "id": "claude_3_opus_inspect", "releaseDate": "2024-03-04", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 3.952262, "ci_low": 1.706313, "ci_high": 8.76484 }, "p80": { "estimate": 0.638973, "ci_low": 0.190491, "ci_high": 2.118276 } }, { "id": "gpt_4_1106_inspect", "releaseDate": "2023-11-06", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 4.044959, "ci_low": 1.866859, "ci_high": 8.443226 }, "p80": { "estimate": 0.783032, "ci_low": 0.276599, "ci_high": 2.358414 } }, { "id": "gpt_4", "releaseDate": "2023-03-14", "scaffolds": [ "modular-public", null ], "p50": { "estimate": 3.987428, "ci_low": 1.93292, "ci_high": 7.995283 }, "p80": { "estimate": 0.889561, "ci_low": 0.342984, "ci_high": 2.523746 } } ], "sha256": "aae31902b0519a4da73e16643915e5e8aca13cd3315c3aac893ce3d6dfe92ad9", "etag": "\"87470c88ac784af063e5d1bb1c58788b-ssl-df\"", "lastModified": null, "retrievedAt": "2026-09-26T05:37:41.148Z", "publishedAt": null, "measuredAt": null }, "previous": { "benchmark": "METR-Horizon-v1.1", "unit": "human-expert minutes", "intervalLevel": 0.95, "longTasksVersion": "799cc9c4b4483a93fc3445623a49ea1bd74fdeb2", "swaaVersion": "f6cc84052e2a79dd540766c8a1b15ff399696371", "excludedLegacy": 3, "records": [ { "id": "claude_mythos_preview_early_inspect", "releaseDate": "2026-04-07", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 1044.780145, "ci_low": 508.876789, "ci_high": 3304.261235 }, "p80": { "estimate": 185.911829, "ci_low": 97.30292, "ci_high": 398.514614 } }, { "id": "gpt_5_4", "releaseDate": "2026-03-05", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 341.735276, "ci_low": 186.581591, "ci_high": 768.779526 }, "p80": { "estimate": 53.877851, "ci_low": 23.957027, "ci_high": 108.679232 } }, { "id": "gemini_3_1_pro", "releaseDate": "2026-02-19", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 384.147435, "ci_low": 233.50073, "ci_high": 694.750898 }, "p80": { "estimate": 89.801503, "ci_low": 52.025934, "ci_high": 158.618017 } }, { "id": "claude_opus_4_6_inspect", "releaseDate": "2026-02-05", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 718.80683, "ci_low": 316.685725, "ci_high": 3633.786163 }, "p80": { "estimate": 69.874587, "ci_low": 27.026521, "ci_high": 170.437873 } }, { "id": "gpt_5_3_codex", "releaseDate": "2026-02-05", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 349.530732, "ci_low": 194.913134, "ci_high": 816.351994 }, "p80": { "estimate": 54.739407, "ci_low": 22.361504, "ci_high": 122.058963 } }, { "id": "gpt_5_2", "releaseDate": "2025-12-11", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 352.249302, "ci_low": 198.067494, "ci_high": 815.177445 }, "p80": { "estimate": 66.002649, "ci_low": 31.665111, "ci_high": 131.730719 } }, { "id": "claude_opus_4_5_inspect", "releaseDate": "2025-11-24", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 292.994594, "ci_low": 161.717714, "ci_high": 623.704698 }, "p80": { "estimate": 49.430584, "ci_low": 20.660485, "ci_high": 104.641318 } }, { "id": "gpt_5_1_codex_max_inspect", "releaseDate": "2025-11-19", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 223.714694, "ci_low": 134.31077, "ci_high": 396.206723 }, "p80": { "estimate": 50.632499, "ci_low": 27.30204, "ci_high": 88.365435 } }, { "id": "gemini_3_pro", "releaseDate": "2025-11-18", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 224.325884, "ci_low": 139.565157, "ci_high": 379.235434 }, "p80": { "estimate": 54.142849, "ci_low": 25.650451, "ci_high": 103.295501 } }, { "id": "gpt_5_2025_08_07_inspect", "releaseDate": "2025-08-07", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 203.012577, "ci_low": 112.641357, "ci_high": 405.551565 }, "p80": { "estimate": 38.312431, "ci_low": 19.184874, "ci_high": 70.093125 } }, { "id": "claude_4_1_opus_inspect", "releaseDate": "2025-08-05", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 100.472004, "ci_low": 59.272391, "ci_high": 159.451918 }, "p80": { "estimate": 23.455761, "ci_low": 10.538456, "ci_high": 47.785951 } }, { "id": "claude_4_opus_inspect", "releaseDate": "2025-05-22", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 100.366123, "ci_low": 59.994073, "ci_high": 163.454074 }, "p80": { "estimate": 20.429752, "ci_low": 8.483125, "ci_high": 41.937268 } }, { "id": "o3_inspect", "releaseDate": "2025-04-16", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 119.732634, "ci_low": 74.615398, "ci_high": 190.943818 }, "p80": { "estimate": 29.981603, "ci_low": 15.127348, "ci_high": 57.823511 } }, { "id": "claude_3_7_sonnet_inspect", "releaseDate": "2025-02-24", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 60.388937, "ci_low": 33.006168, "ci_high": 104.226017 }, "p80": { "estimate": 12.09179, "ci_low": 4.575608, "ci_high": 28.889879 } }, { "id": "o1_inspect", "releaseDate": "2024-12-05", "scaffolds": [ "mtb/start_metr_task,triframe_inspect/triframe_agent", "ai_rd_fix_embedding/init_solver,use_tools,triframe_inspect/triframe_agent", "re_bench_common/init_solver,use_tools,triframe_inspect/triframe_agent", "metr_tasks_swaa/init_solver,generate", "triframe_inspect/triframe_agent" ], "p50": { "estimate": 38.831588, "ci_low": 21.164512, "ci_high": 64.95249 }, "p80": { "estimate": 7.090121, "ci_low": 3.036033, "ci_high": 16.640636 } }, { "id": "claude_3_5_sonnet_20241022_inspect", "releaseDate": "2024-10-22", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 20.522872, "ci_low": 10.144026, "ci_high": 40.82028 }, "p80": { "estimate": 2.595677, "ci_low": 0.89379, "ci_high": 7.363864 } }, { "id": "o1_preview", "releaseDate": "2024-09-12", "scaffolds": [ "duet", null ], "p50": { "estimate": 20.326586, "ci_low": 11.716193, "ci_high": 33.379877 }, "p80": { "estimate": 4.420545, "ci_low": 2.012646, "ci_high": 8.912444 } }, { "id": "claude_3_5_sonnet_20240620_inspect", "releaseDate": "2024-06-20", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 11.395377, "ci_low": 5.489734, "ci_high": 22.384214 }, "p80": { "estimate": 1.671757, "ci_low": 0.57927, "ci_high": 4.573519 } }, { "id": "gpt_4o_inspect", "releaseDate": "2024-05-13", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 6.991195, "ci_low": 4.001482, "ci_high": 12.905741 }, "p80": { "estimate": 1.267009, "ci_low": 0.562997, "ci_high": 3.01517 } }, { "id": "gpt_4_turbo_inspect", "releaseDate": "2024-04-09", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 3.732787, "ci_low": 1.980046, "ci_high": 6.736613 }, "p80": { "estimate": 0.927933, "ci_low": 0.428277, "ci_high": 2.196806 } }, { "id": "claude_3_opus_inspect", "releaseDate": "2024-03-04", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 3.952262, "ci_low": 1.706313, "ci_high": 8.76484 }, "p80": { "estimate": 0.638973, "ci_low": 0.190491, "ci_high": 2.118276 } }, { "id": "gpt_4_1106_inspect", "releaseDate": "2023-11-06", "scaffolds": [ "mtb/start_metr_task,metr_agents/react", "ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react", "re_bench_common/init_solver,use_tools,metr_agents/react", "metr_tasks_swaa/init_solver,generate", "metr_agents/react" ], "p50": { "estimate": 4.044959, "ci_low": 1.866859, "ci_high": 8.443226 }, "p80": { "estimate": 0.783032, "ci_low": 0.276599, "ci_high": 2.358414 } }, { "id": "gpt_4", "releaseDate": "2023-03-14", "scaffolds": [ "modular-public", null ], "p50": { "estimate": 3.987428, "ci_low": 1.93292, "ci_high": 7.995283 }, "p80": { "estimate": 0.889561, "ci_low": 0.342984, "ci_high": 2.523746 } } ], "sha256": "aae31902b0519a4da73e16643915e5e8aca13cd3315c3aac893ce3d6dfe92ad9", "etag": "\"87470c88ac784af063e5d1bb1c58788b-ssl-df\"", "lastModified": null, "retrievedAt": "2026-09-24T21:05:55.895Z", "publishedAt": null, "measuredAt": null }, "changeSummary": "Since collection 2026-09-24T21:05:55.895Z: 0 added, 0 removed, 0 revised model measurements; 0 changed scaffolds (not compared).", "context": { "id": "2026-0", "textSha256": "88410c0c14200538a49389293abf03e5d3412e0c5eb6f8a804cdb234f0a8cfd8", "forecastSha256": "662a4369b05520d70255c65d8a52da1eee57923ffc0d171d53e3e08c234a568a", "role": "Context only: software-task capability, not human-reviewed workflows, monitored runtime or safeguards. No forecast milestone is assessed." } } }, "forecastVersion": { "schemaVersion": 1, "sha256": "662a4369b05520d70255c65d8a52da1eee57923ffc0d171d53e3e08c234a568a" }, "updated": "2026-09-26T21:15:54.484Z", "note": "14 predictions carry one reviewed item of direct evidence: an authoritative news article fetched and quote-checked at publication, labeled direct, scenario or leading-indicator, and recorded with what it does not establish. 46 predictions have no qualifying source published inside the 14-day window and are recorded as uncited rather than evidenced by something weaker. Reuse is restricted to reviewed compatible concept families; a source cited by more than one prediction is counted once.", "source": "news-verified", "sourceStatus": { "mode": "news-verified", "primarySource": "live-verified-news", "activeSource": "news-verified", "reason": "x-evidence-retired-2026-08-13", "message": "Predictions are evidenced only by authoritative news and research published inside the 14-day currency window. Every citation is fetched live at review and again at publish, its verbatim quote re-checked in the fetched text, and a SHA-256 of that text compared against the reviewed hash. A prediction with no qualifying source in the window is recorded as uncited rather than left blank.", "actionRequired": null, "retiredSources": [ "x-api", "archive-verified" ], "httpStatus": null, "windowDays": 14, "verificationPaceMs": null }, "sourceAttempts": [ { "source": "x-api", "status": "retired", "count": 0, "reason": "x-evidence-retired-2026-08-13", "detail": "The X API and the @peterxing archive corpus were retired at the site owner's instruction. Predictions are now evidenced only by live-verified news published inside the currency window." }, { "source": "archive-verified", "status": "retired", "count": 0, "reason": "x-evidence-retired-2026-08-13" } ], "sourceFetchedAt": "2026-09-26T21:15:25.481Z", "sourceFresh": true, "newestItemAt": "2026-09-26T16:34:59.000Z", "uncited": { "windowDays": 14, "count": 46, "items": { "2026-5": { "id": "2026-5", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.472Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2027-0": { "id": "2027-0", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.473Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2027-2": { "id": "2027-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.473Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2027-4": { "id": "2027-4", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.473Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2028-7": { "id": "2028-7", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2029-0": { "id": "2029-0", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2029-3": { "id": "2029-3", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2029-5": { "id": "2029-5", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2030-1": { "id": "2030-1", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2030-2": { "id": "2030-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2031-0": { "id": "2031-0", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2031-1": { "id": "2031-1", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2031-2": { "id": "2031-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2032-0": { "id": "2032-0", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2032-2": { "id": "2032-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.474Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2032-4": { "id": "2032-4", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2032-5": { "id": "2032-5", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2033-0": { "id": "2033-0", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2033-1": { "id": "2033-1", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2033-2": { "id": "2033-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2033-4": { "id": "2033-4", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2033-5": { "id": "2033-5", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2034-1": { "id": "2034-1", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2034-3": { "id": "2034-3", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2034-4": { "id": "2034-4", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2035-2": { "id": "2035-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2035-3": { "id": "2035-3", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2035-4": { "id": "2035-4", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2036-2": { "id": "2036-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2036-3": { "id": "2036-3", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2036-4": { "id": "2036-4", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2036-5": { "id": "2036-5", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.475Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2037-2": { "id": "2037-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2037-4": { "id": "2037-4", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2037-5": { "id": "2037-5", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2038-3": { "id": "2038-3", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2038-4": { "id": "2038-4", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2039-0": { "id": "2039-0", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2039-2": { "id": "2039-2", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2039-3": { "id": "2039-3", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2039-5": { "id": "2039-5", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "2040-1": { "id": "2040-1", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "horizon-whole-brain-emulation-and-uploading": { "id": "horizon-whole-brain-emulation-and-uploading", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "horizon-kardashev-energy-scaling": { "id": "horizon-kardashev-energy-scaling", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "horizon-transcension-hypothesis": { "id": "horizon-transcension-hypothesis", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." }, "horizon-ruliad-testable-physics": { "id": "horizon-ruliad-testable-physics", "reason": "no-qualifying-source-in-window", "windowDays": 14, "searchedAt": "2026-09-26T21:15:54.476Z", "statement": "No authoritative source published in the last 14 days was found for this prediction. Nothing older, adjacent or unreviewed has been substituted." } } }, "context": { "windowDays": 14, "count": 43, "items": { "2026-0": { "id": "news:microsoft-computer-use-agent-worlds", "sourceKey": "https://www.microsoft.com/en-us/research/blog/echoverse-deep-evolving-environments-for-computer-use-agents", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Microsoft Research", "publisherHost": "microsoft.com", "byline": "alyssa", "headline": "Deep, evolving environments for computer-use agents", "quote": "Agents often struggle with the same challenging UI elements, like date pickers and nested filters.", "text": "Agents often struggle with the same challenging UI elements, like date pickers and nested filters.", "url": "https://www.microsoft.com/en-us/research/blog/echoverse-deep-evolving-environments-for-computer-use-agents/", "articleDate": "2026-07-30T17:00:00.000Z", "publishedAt": "2026-07-30T17:00:00.000Z", "publishedAtSource": "page", "ageDays": 58, "ageBucket": "31-90d", "windowDays": 14, "date": "30 Jul 2026", "sourceQuality": "official-research-organization", "evidenceType": "leading-indicator", "mappingRationale": "Microsoft Research describes purpose-built training and evaluation worlds for computer-use agents and records that agents still struggle with ordinary interface elements. It evidences the industrial effort to make agent workflows reliable and the state of the art; it does not evidence multi-hour reliability, day- or week-long research runs, or trajectory-level safeguards.", "reuseFamily": "computer-use-agents", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Frontier agents reliably complete multi-hour computer workflows with human review, while monitored research systems can persist for days or weeks and require trajectory-level safeguards", "provenance": { "evidenceOwner": "news", "publisher": "Microsoft Research", "publisherHost": "microsoft.com", "byline": "alyssa", "publishedAt": "2026-07-30T17:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "official-research-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "48e66aedf1cfdbc65d5bca30a827e6ca6e59544261abeda33c6a0511e93f272d" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 58 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2026-2": { "id": "news:cnbc-ai-lab-lobbying-gap", "sourceKey": "https://www.cnbc.com/2026/07/21/openai-anthropic-ai-lobbying-spending-q2-2026.html", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "CNBC", "publisherHost": "cnbc.com", "byline": null, "headline": "OpenAI, Anthropic boost lobbying as legacy tech and defense spending slips", "quote": "Established technology and defense companies still spend far more overall, but OpenAI and Anthropic are narrowing the gap with some of Washington's biggest corporate lobbying operations.", "text": "Established technology and defense companies still spend far more overall, but OpenAI and Anthropic are narrowing the gap with some of Washington's biggest corporate lobbying operations.", "url": "https://www.cnbc.com/2026/07/21/openai-anthropic-ai-lobbying-spending-q2-2026.html", "articleDate": "2026-07-21T16:30:11.000Z", "publishedAt": "2026-07-21T16:30:11.000Z", "publishedAtSource": "page", "ageDays": 67, "ageBucket": "31-90d", "windowDays": 14, "date": "21 Jul 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "CNBC reports federal lobbying disclosures showing two frontier AI labs narrowing the gap with the largest established corporate lobbying operations in Washington. It evidences the political-leverage half of the prediction through disclosed spending; it does not evidence any trillion-dollar valuation.", "reuseFamily": "frontier-lab-political-power", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Frontier AI labs and infrastructure firms accumulate trillion-dollar-scale valuations and exceptional political leverage", "provenance": { "evidenceOwner": "news", "publisher": "CNBC", "publisherHost": "cnbc.com", "byline": null, "publishedAt": "2026-07-21T16:30:11.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "f98b909c3fade7ab06b527fb2342b6b4fc3ca08942067982a968fe8d7c273de5" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 67 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2026-3": { "id": "news:ieee-persona-humanoid-welding", "sourceKey": "https://spectrum.ieee.org/persona-ai-humanoid-robot-welding", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "headline": "Inside Persona’s Bold Bet On Humanoid Welders In Shipyards", "quote": "These are the same environments with the same sorts of potential applications that basically every other humanoid robotics company is attempting to make economically viable, and despite an ever more exhaustive number of demonstrations, so far none have succeeded at any sort of useful scale.", "text": "These are the same environments with the same sorts of potential applications that basically every other humanoid robotics company is attempting to make economically viable, and despite an ever more exhaustive number …", "url": "https://spectrum.ieee.org/persona-ai-humanoid-robot-welding", "articleDate": "2026-08-17T15:33:42.000Z", "publishedAt": "2026-08-17T15:33:42.000Z", "publishedAtSource": "page", "ageDays": 40, "ageBucket": "31-90d", "windowDays": 14, "date": "17 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "The prediction is two-sided: humanoids move onto live factory lines in the thousands BUT remain far short of general physical labor. This reported IEEE Spectrum account of Persona AI's shipyard-welding programme, with named industrial partners, evidences the second half directly and in the industry's own terms — every humanoid company is attempting to make the same environments economically viable and, despite an ever-growing number of demonstrations, none has succeeded at any useful scale, which is why Persona deliberately narrowed to a single robot-friendly skilled task. It also evidences the entry of humanoids into economically valuable industrial work. It does NOT evidence that thousands of humanoids are on live factory lines: the article reports a customer-scale ambition of hundreds of robots per location, and an ambition is not deployed capacity. This mapping stands on its own positive case for 2026-3 and is not a relocation of the separately rejected 2032-1 proposal.", "reuseFamily": "humanoid-industrial-deployment", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-17", "lastVerifiedAt": "2026-08-17", "maps": "Humanoid robots move onto live factory lines in the thousands, but remain far short of general physical labor", "provenance": { "evidenceOwner": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "publishedAt": "2026-08-17T15:33:42.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-17", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-17", "textSha256": "0e441d70fd87f0847123c3cf70eecf13577b9dc1a35303f249178ad7a29237ad" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 40 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2026-4": { "id": "news:ec-ai-act-enforcement-august", "sourceKey": "https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Shaping Europe’s digital future", "publisherHost": "europa.eu", "byline": null, "headline": "Commission starts enforcing AI Act rules and new transparency requirements on 2 August", "quote": "From 2 August 2026, the European Commission’s AI Office, together with national authorities, will begin enforcing the Artificial Intelligence (AI) Act.", "text": "From 2 August 2026, the European Commission’s AI Office, together with national authorities, will begin enforcing the Artificial Intelligence (AI) Act.", "url": "https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august", "articleDate": "2026-08-02T00:00:00.000Z", "publishedAt": "2026-08-02T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 56, "ageBucket": "31-90d", "windowDays": 14, "date": "02 Aug 2026", "sourceQuality": "intergovernmental-organization", "evidenceType": "direct", "mappingRationale": "The European Commission records the AI Office and national authorities beginning enforcement of the AI Act, evidencing the EU half of the prediction. It concerns transparency obligations and does not evidence US practice or cyber, bio and autonomy release thresholds.", "reuseFamily": "frontier-governance", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-13", "lastVerifiedAt": "2026-08-13", "maps": "Frontier-model release review becomes standard in the US and EU for cyber, bio and autonomy thresholds", "provenance": { "evidenceOwner": "news", "publisher": "Shaping Europe’s digital future", "publisherHost": "europa.eu", "byline": null, "publishedAt": "2026-08-02T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-13", "sourceQuality": "intergovernmental-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-13", "textSha256": "48cda069c6afe2e42f715ee6d96b89eab7e1dbc34aaffd1f9ad7ed29e8df889e" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 56 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2026-6": { "id": "news:guardian-astra-release-contested-agi-claim", "sourceKey": "https://www.theguardian.com/technology/2026/sep/03/openai-artificial-general-intelligence-astra-release", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": null, "headline": "OpenAI hails ‘new era of artificial general intelligence’ with Astra model release", "quote": "The president of OpenAI, Greg Brockman, has claimed the world has entered a new era of artificial general intelligence after the release of his company’s latest model, Astra, which it described as the “world’s most intelligent and aligned model”.", "text": "The president of OpenAI, Greg Brockman, has claimed the world has entered a new era of artificial general intelligence after the release of his company’s latest model, Astra, which it described as the “world’s most in…", "url": "https://www.theguardian.com/technology/2026/sep/03/openai-artificial-general-intelligence-astra-release", "articleDate": "2026-09-03T18:24:42.000Z", "publishedAt": "2026-09-03T18:24:42.000Z", "publishedAtSource": "page", "ageDays": 23, "ageBucket": "15-30d", "windowDays": 14, "date": "03 Sept 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "The Guardian reports the actual release of Astra and attributes an AGI-era claim to OpenAI president Greg Brockman. This bears directly on the forecast subject and the shipping facet, not generic AI adjacency. IT DOES NOT ESTABLISH GENUINE HUMAN-LEVEL AGI: the article describes AGI as a fuzzy threshold, reports differing definitions, and quotes OpenAI CEO Sam Altman calling it poorly defined and an irrelevant marketing term. Capability and benchmark assertions are company claims, not an independently reproduced cross-domain human-level evaluation. The cyber incidents discussed involved other models, not Astra. This is a narrowly scoped release/claim indicator with explicit uncertainty, not independent AGI confirmation, an on-track verdict, a resolved forecast or a reason to change its probability. A release and an AGI label cannot establish the genuine-capability threshold.", "reuseFamily": "contested-human-level-agi-release", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-09-12T12:57:00.966Z", "lastVerifiedAt": "2026-09-12T23:01:39.806Z", "maps": "Signal: my call — genuine human-level AGI ships by end of 2026", "provenance": { "evidenceOwner": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": null, "publishedAt": "2026-09-03T18:24:42.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-12T23:01:39.806Z", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-12T23:01:39.806Z", "textSha256": "2d822237fa1c8582f1e359beccc088c7b630deb67290250db9cc18ca2a504812" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 23 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2026-7": { "id": "news:techreview-bci-trials-taking-off", "sourceKey": "https://www.technologyreview.com/2026/06/19/1139270/brain-computer-interface-trials-are-taking-off", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Jessica Hamzelou", "headline": "Brain-computer interface trials are taking off", "quote": "He has now spent almost three years using a brain-computer interface (BCI) that enables him to “speak,” surf the web, and perform his job as a climate activist, largely independently.", "text": "He has now spent almost three years using a brain-computer interface (BCI) that enables him to “speak,” surf the web, and perform his job as a climate activist, largely independently.", "url": "https://www.technologyreview.com/2026/06/19/1139270/brain-computer-interface-trials-are-taking-off/", "articleDate": "2026-06-19T09:00:00.000Z", "publishedAt": "2026-06-19T09:00:00.000Z", "publishedAtSource": "page", "ageDays": 100, "ageBucket": "91-365d", "windowDays": 14, "date": "19 Jun 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "Reports a single participant using an intracortical BCI for speech and computer control for almost three years, largely independently — the multi-year home-use regime the prediction thresholds on. It does not state the 3,800-hour figure, does not report a peer-reviewed hour count, and says nothing about Neuralink PRIME-family trial posture.", "reuseFamily": "bci-home-use", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Peer-reviewed intracortical BCI home use surpasses 3,800 hours for speech and cursor control in one participant; Neuralink PRIME-family trials remain safety-first", "provenance": { "evidenceOwner": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Jessica Hamzelou", "publishedAt": "2026-06-19T09:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "30170e5c7fbba96ef291ed5344a07285373855715e4120e30a408a594fecf7be" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 100 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2026-8": { "id": "news:ars-orbital-datacenter-constraints-1", "sourceKey": "https://arstechnica.com/space/2026/07/how-hard-is-it-to-build-orbital-data-centers-actually", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Eric Berger", "headline": "How hard is it to build orbital data centers, actually?", "quote": "The spacecraft is due to launch in October and, if successful, will demonstrate the ability to radiate heat efficiently and run useful workloads for customers, Johnston said.", "text": "The spacecraft is due to launch in October and, if successful, will demonstrate the ability to radiate heat efficiently and run useful workloads for customers, Johnston said.", "url": "https://arstechnica.com/space/2026/07/how-hard-is-it-to-build-orbital-data-centers-actually/", "articleDate": "2026-07-15T11:00:09.000Z", "publishedAt": "2026-07-15T11:00:09.000Z", "publishedAtSource": "page", "ageDays": 73, "ageBucket": "31-90d", "windowDays": 14, "date": "15 Jul 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "The prediction thresholds on orbital compute remaining demonstrator-scale through 2026 with no operator disclosing utility-scale power and cooling. This reported analysis describes the next flight as a 450 kg satellite with 8 kW of generation that has yet to demonstrate efficient heat rejection or customer workloads — demonstrator scale, stated by the operator. It evidences the state of the art; it does not evidence that the 2026 outcome has occurred.", "reuseFamily": "orbital-compute-constraints", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-17", "lastVerifiedAt": "2026-08-17", "maps": "Orbital compute remains demonstrator-scale through 2026: launched GPUs run named AI workloads, but no operator discloses utility-scale power and cooling", "provenance": { "evidenceOwner": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Eric Berger", "publishedAt": "2026-07-15T11:00:09.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-17", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-17", "textSha256": "a2decd2e64052479bc0e9c6a647fff54a847d230cb63649be05c3b0054969762" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 73 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "aws-waf x-amzn-waf-action: captcha (HTTP 405)", "lastCheckedAt": "2026-09-26T21:15:33.403Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } }, "2027-1": { "id": "news:ars-coding-agents-burnout-limits", "sourceKey": "https://arstechnica.com/information-technology/2026/01/10-things-i-learned-from-burning-myself-out-with-ai-coding-agents", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Benj Edwards", "headline": "10 things I learned from burning myself out with AI coding agents", "quote": "Even with that impression, though, I know these are hobby projects, and the limitations of coding agents lead me to believe that veteran software developers probably shouldn’t fear losing their jobs to these tools any time soon.", "text": "Even with that impression, though, I know these are hobby projects, and the limitations of coding agents lead me to believe that veteran software developers probably shouldn’t fear losing their jobs to these tools any…", "url": "https://arstechnica.com/information-technology/2026/01/10-things-i-learned-from-burning-myself-out-with-ai-coding-agents/", "articleDate": "2026-01-19T12:00:45.000Z", "publishedAt": "2026-01-19T12:00:45.000Z", "publishedAtSource": "page", "ageDays": 250, "ageBucket": "91-365d", "windowDays": 14, "date": "19 Jan 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "A named Ars Technica journalist reports sustained hands-on use of AI coding agents and concludes their present limitations mean veteran developers should not expect to be replaced soon. It evidences the current state of end-to-end AI software production and, on its face, that the prediction has NOT yet occurred; it is not evidence that it will.", "reuseFamily": "ai-software-production", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Most production software is written end-to-end by AI, with humans specifying goals and reviewing exceptions", "provenance": { "evidenceOwner": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Benj Edwards", "publishedAt": "2026-01-19T12:00:45.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "72dec099aa537b94a2e6a42cb101f0a2b0ee637c934aead263c7d690736dbe80" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 250 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "aws-waf x-amzn-waf-action: captcha (HTTP 405)", "lastCheckedAt": "2026-09-26T21:15:33.950Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } }, "2027-3": { "id": "news:ars-meta-ai-native-workforce-redesign", "sourceKey": "https://arstechnica.com/ai/2026/08/metas-scrapped-plans-to-go-ai-native-included-slashing-teams-by-60-percent", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Scharon Harding", "headline": "AI agents meant to replace Meta workers made “large-scale, disruptive actions”", "quote": "Meta’s case suggests that even some of the most eager organizations may struggle to replace various human workloads with AI while also highlighting the risks of overzealous AI projects.", "text": "Meta’s case suggests that even some of the most eager organizations may struggle to replace various human workloads with AI while also highlighting the risks of overzealous AI projects.", "url": "https://arstechnica.com/ai/2026/08/metas-scrapped-plans-to-go-ai-native-included-slashing-teams-by-60-percent/", "articleDate": "2026-08-26T21:25:27.000Z", "publishedAt": "2026-08-26T21:25:27.000Z", "publishedAtSource": "page", "ageDays": 31, "ageBucket": "31-90d", "windowDays": 14, "date": "26 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "Ars Technica reports Meta's internal exercise to go 'AI-native', including scenarios that would have cut teams by up to 60 percent, and the large-scale disruptive actions its AI agents took during that exercise. A major firm modelling the wholesale replacement of human workloads with AI agents is exactly the simultaneous business-model and workforce redesign this prediction describes, and it is reported first-hand rather than forecast. IT DOES NOT EVIDENCE THE PREDICTION. It concerns ONE company in ONE industry, not disruption 'across every major industry'; the plans were scrapped rather than implemented, so no workforce redesign actually occurred; and the quoted assessment cuts against the prediction's pace by concluding that even the most eager organizations may STRUGGLE to replace human workloads with AI. It also makes no claim that the systems involved are human-level.", "reuseFamily": "ai-workforce-redesign", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-09-03", "lastVerifiedAt": "2026-09-03", "maps": "Human-level AI becomes operationally disruptive across every major industry, forcing simultaneous business-model and workforce redesign", "provenance": { "evidenceOwner": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Scharon Harding", "publishedAt": "2026-08-26T21:25:27.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-03", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-03", "textSha256": "a04b7d226b3e3a569818d13bbffb2809dd8f69fd808b30d912a6ae576f65b5a8" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 31 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "aws-waf x-amzn-waf-action: captcha (HTTP 405)", "lastCheckedAt": "2026-09-26T21:15:34.347Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } }, "2027-5": { "id": "news:constructiondive-datacentre-power-political-delays", "sourceKey": "https://www.constructiondive.com/news/data-center-project-cancellations-power-public-pushback/818157", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Construction Dive", "publisherHost": "constructiondive.com", "byline": "Sebastian Obando", "headline": "What’s stalling data center projects? Public opposition and power access lead delays.", "quote": "The projects’ need for huge amounts of power has created obstacles to build and increased prices. Meanwhile, communities and political groups have opposed the projects, leading to delays or abandonments.", "text": "The projects’ need for huge amounts of power has created obstacles to build and increased prices. Meanwhile, communities and political groups have opposed the projects, leading to delays or abandonments.", "url": "https://www.constructiondive.com/news/data-center-project-cancellations-power-public-pushback/818157/", "articleDate": "2026-04-22T00:00:00.000Z", "publishedAt": "2026-04-22T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 158, "ageBucket": "91-365d", "windowDays": 14, "date": "22 Apr 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "Construction Dive reports concrete power-access, utility-timing and political/permitting obstacles to data center construction, with named developer interviews and an attributed cancellation count. This bears directly on the power/grid and political-constraint facets of the forecast. It does not measure water scarcity, establish a worldwide ranking of constraints, or resolve the 2027 forecast.", "reuseFamily": "datacentre-energy", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-09-21T22:56:50.086Z", "lastVerifiedAt": "2026-09-21T22:56:50.086Z", "maps": "Datacenter power, water and grid capacity become top-tier infrastructure and political constraints", "provenance": { "evidenceOwner": "news", "publisher": "Construction Dive", "publisherHost": "constructiondive.com", "byline": "Sebastian Obando", "publishedAt": "2026-04-22T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-21T22:56:50.086Z", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-21T22:56:50.086Z", "textSha256": "40bf3e0f48f355e7abc1714eeaa80479bea91b2d9499f0695ef9e639fe743fee" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 158 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2028-0": { "id": "news:guardian-au-ai-law-election-risk", "sourceKey": "https://www.theguardian.com/australia-news/2026/aug/25/albanese-seeks-to-quell-datacentre-disquiet-as-climate-expert-warns-weve-got-one-shot-to-get-the-rules-right", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": null, "headline": "Albanese seeks to quell datacentre disquiet as climate expert warns ‘we’ve got one shot to get the rules right’", "quote": "Combining the elements into a single piece of legislation signals the government’s ambition on AI, but could also heighten the political risk of getting the bill through parliament ahead of the next election.", "text": "Combining the elements into a single piece of legislation signals the government’s ambition on AI, but could also heighten the political risk of getting the bill through parliament ahead of the next election.", "url": "https://www.theguardian.com/australia-news/2026/aug/25/albanese-seeks-to-quell-datacentre-disquiet-as-climate-expert-warns-weve-got-one-shot-to-get-the-rules-right", "articleDate": "2026-08-24T14:01:35.000Z", "publishedAt": "2026-08-24T14:01:35.000Z", "publishedAtSource": "page", "ageDays": 33, "ageBucket": "31-90d", "windowDays": 14, "date": "24 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "The Guardian reports the Australian prime minister taking a single omnibus AI bill to national cabinet against opposition from state governments, and states explicitly that combining the measures heightens the political risk of passing it before the next election. It evidences AI policy becoming a contested national electoral issue in one democracy; it does not evidence that AI became the LARGEST issue in any election, it concerns a legislative fight rather than a campaign, and it says nothing about 2028.", "reuseFamily": "ai-electoral-politics", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-25", "lastVerifiedAt": "2026-08-25", "maps": "AI becomes the largest issue in a major national election", "provenance": { "evidenceOwner": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": null, "publishedAt": "2026-08-24T14:01:35.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-25", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-25", "textSha256": "0327ac7cb1befb45b71b257a313ae7d22ffa0823b174b9ae0e2569f57c588fc5" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 33 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2028-1": { "id": "news:challenger-ai-labour-market", "sourceKey": "https://www.challengergray.com/blog/challenger-report-layoffs-fall-hiring-picks-up-ai-leads-for-fifth-straight-month", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Challenger, Gray & Christmas, Inc. | Outplacement & Career Transitioning Services", "publisherHost": "challengergray.com", "byline": "Colleen Madden Blumenfeld", "headline": "Challenger Report: Layoffs Fall, Hiring Picks Up; AI Leads For Fifth Straight Month", "quote": "“Hiring has also increased over last year by 25%, so while AI is shifting the labor market, it is not dismantling it,” said Andy Challenger, workplace expert and chief revenue officer for Challenger, Gray & Christmas.", "text": "“Hiring has also increased over last year by 25%, so while AI is shifting the labor market, it is not dismantling it,” said Andy Challenger, workplace expert and chief revenue officer for Challenger, Gray & Christmas.", "url": "https://www.challengergray.com/blog/challenger-report-layoffs-fall-hiring-picks-up-ai-leads-for-fifth-straight-month/", "articleDate": "2026-08-06T09:30:00.000Z", "publishedAt": "2026-08-06T09:30:00.000Z", "publishedAtSource": "page", "ageDays": 51, "ageBucket": "31-90d", "windowDays": 14, "date": "06 Aug 2026", "sourceQuality": "named-expert-analysis", "evidenceType": "leading-indicator", "mappingRationale": "Challenger, Gray & Christmas attribute the leading stated reason for US layoffs to AI for a fifth consecutive month while hiring rose 25%, evidencing AI reshaping white-collar work. It does not evidence that most professions yet supervise AI agents.", "reuseFamily": "ai-labour-market", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-13", "lastVerifiedAt": "2026-08-13", "maps": "Most white-collar professions in leading economies revolve around supervising and coordinating AI agents", "provenance": { "evidenceOwner": "news", "publisher": "Challenger, Gray & Christmas, Inc. | Outplacement & Career Transitioning Services", "publisherHost": "challengergray.com", "byline": "Colleen Madden Blumenfeld", "publishedAt": "2026-08-06T09:30:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-13", "sourceQuality": "named-expert-analysis", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-13", "textSha256": "65c1f4a1c9375966e5d9af23a0db47d185c8aad83c9b0f01d8765f2985ef5d80" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 51 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2028-2": { "id": "news:guardian-hollywood-ai-training-agencies", "sourceKey": "https://www.theguardian.com/technology/2026/aug/22/the-hollywood-creatives-training-ai-to-do-their-jobs", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": null, "headline": "‘Digging the grave of my profession’: the Hollywood creatives training AI to do their jobs", "quote": "With feelings ranging from fatalism to guilt, the creatives have signed up with some of the booming training agencies, which have contracts with the biggest AI companies including Anthropic and OpenAI, to pass on hard-won human skills in industries such as finance, health, law and social work.", "text": "With feelings ranging from fatalism to guilt, the creatives have signed up with some of the booming training agencies, which have contracts with the biggest AI companies including Anthropic and OpenAI, to pass on hard…", "url": "https://www.theguardian.com/technology/2026/aug/22/the-hollywood-creatives-training-ai-to-do-their-jobs", "articleDate": "2026-08-22T06:00:55.000Z", "publishedAt": "2026-08-22T06:00:55.000Z", "publishedAtSource": "page", "ageDays": 36, "ageBucket": "31-90d", "windowDays": 14, "date": "22 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "The Guardian reports specialised training agencies under contract to frontier labs recruiting working professionals to transfer their craft into AI systems. That is the profession-by-profession expert-interview pipeline the prediction describes, observed in one industry; it does not evidence the industrialisation of that pipeline across professions, nor the use of deployment data or trained environments.", "reuseFamily": "profession-training-pipelines", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Frontier labs industrialize profession-by-profession training using expert interviews, environments and deployment data", "provenance": { "evidenceOwner": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": null, "publishedAt": "2026-08-22T06:00:55.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "2749302f202fec3c51aa385182d14497448757e55b47daa4615eb21a8b87c6c2" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 36 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2028-3": { "id": "news:ars-deepseek-export-controls-chips", "sourceKey": "https://arstechnica.com/ai/2026/07/facing-us-export-controls-chinas-deepseek-plans-to-make-its-own-chips", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Samuel Axon", "headline": "Facing US export controls, China's DeepSeek plans to make its own chips", "quote": "Huawei controls about half of the data center chip market there, and DeepSeek isn’t the only one trying to enter; Chinese tech giants like Alibaba and Baidu have been making moves, too.", "text": "Huawei controls about half of the data center chip market there, and DeepSeek isn’t the only one trying to enter; Chinese tech giants like Alibaba and Baidu have been making moves, too.", "url": "https://arstechnica.com/ai/2026/07/facing-us-export-controls-chinas-deepseek-plans-to-make-its-own-chips/", "articleDate": "2026-07-07T16:14:53.000Z", "publishedAt": "2026-07-07T16:14:53.000Z", "publishedAtSource": "page", "ageDays": 81, "ageBucket": "31-90d", "windowDays": 14, "date": "07 Jul 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "Reports one firm holding about half of China’s data-centre chip market with a handful of named Chinese giants contesting the rest, under US export controls — concentration of frontier AI capability around a few companies and state policy. It evidences the Chinese half and the export-control lever; it does not evidence US corporate concentration or any named head of state or party leader exercising that control.", "reuseFamily": "compute-geopolitics", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Control over frontier AI concentrates around a handful of US and Chinese companies, presidents and party leaders", "provenance": { "evidenceOwner": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Samuel Axon", "publishedAt": "2026-07-07T16:14:53.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "5709998886b94c0dcb56d83a6b5286f896899a083f665a3e5922f019068020be" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 81 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "aws-waf x-amzn-waf-action: captcha (HTTP 405)", "lastCheckedAt": "2026-09-26T21:15:35.328Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } }, "2028-4": { "id": "news:constructiondive-saline-datacentre-project-financing", "sourceKey": "https://www.constructiondive.com/news/walbridge-breaks-ground-stargate-data-center-openai-oracle/821972", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Construction Dive", "publisherHost": "constructiondive.com", "byline": "Sebastian Obando", "headline": "Walbridge breaks ground on $16B Stargate data center", "quote": "In April, Related Digital announced financing had been secured for the $16 billion data center campus project, according to a news release.", "text": "In April, Related Digital announced financing had been secured for the $16 billion data center campus project, according to a news release.", "url": "https://www.constructiondive.com/news/walbridge-breaks-ground-stargate-data-center-openai-oracle/821972/", "articleDate": "2026-06-04T00:00:00.000Z", "publishedAt": "2026-06-04T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 115, "ageBucket": "91-365d", "windowDays": 14, "date": "04 Jun 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "In its June 4, 2026 report, Construction Dive attributes financing for one $16 billion Saline data center campus to Related Digital's April announcement and construction underway on all three halls to Oracle. These are reported company financing/progress claims, not independently audited expenditures or a statement of progress today. This is a project-level capital-commitment leading indicator tied to an identified construction project, not a broad AI-capex total. The announced campus value is not an audited construction-only cost breakdown; it may cover costs beyond buildings. The source does not provide annual aggregate construction commitments, a comparable US defense-budget figure, or evidence that the 2028 threshold has been crossed. No annualization, summation with other announcements, or probability change is inferred.", "reuseFamily": "datacentre-energy", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-09-21T22:56:51.858Z", "lastVerifiedAt": "2026-09-21T22:56:51.858Z", "maps": "Annual datacenter construction commitments exceed the US defense budget", "provenance": { "evidenceOwner": "news", "publisher": "Construction Dive", "publisherHost": "constructiondive.com", "byline": "Sebastian Obando", "publishedAt": "2026-06-04T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-21T22:56:51.858Z", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-21T22:56:51.858Z", "textSha256": "443f8abd9448d97189e8cc8ef2c562df7e95292d8e2251cc7458fe6a660f675a" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 115 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2028-5": { "id": "news:techreview-recursive-self-improvement-timing", "sourceKey": "https://www.technologyreview.com/2026/08/18/1142188/ai-recursive-self-improvement", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Michelle Kim", "headline": "AI’s recursive self-improvement might not come so quickly after all", "quote": "The researchers behind it found that AI agents are not yet capable of conducting open-ended AI research—free-form investigations that have no clear-cut answers and require judgment and taste, which may be integral to building self-improving AI.", "text": "The researchers behind it found that AI agents are not yet capable of conducting open-ended AI research—free-form investigations that have no clear-cut answers and require judgment and taste, which may be integral to …", "url": "https://www.technologyreview.com/2026/08/18/1142188/ai-recursive-self-improvement/", "articleDate": "2026-08-18T09:00:00.000Z", "publishedAt": "2026-08-18T09:00:00.000Z", "publishedAtSource": "page", "ageDays": 40, "ageBucket": "31-90d", "windowDays": 14, "date": "18 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "scenario", "mappingRationale": "The prediction asserts that recursive self-improvement BEGINS on the ungoverned 2028-2030 branch. MIT Technology Review reports a study evaluating whether AI agents can conduct open-ended AI research — the exact mechanism the prediction names — using unpublished NeurIPS submissions so the answers could not be memorised. It finds the agents able to do the engineering but, in a named researcher's words, unambiguously bad at the research itself, and concludes that some hyped timelines for automating AI research may be running ahead of the evidence. THIS SOURCE CUTS AGAINST THE PREDICTION'S TIMING AND IS LABELLED A SCENARIO SOURCE FOR THAT REASON: it is evidence about the state of the mechanism, and it is emphatically NOT evidence that superintelligence has emerged, that recursive self-improvement has begun, or that the 2028-2030 window will be met. It speaks to the AI-R&D-automation facet only and says nothing about the emergence of superintelligence itself.", "reuseFamily": "ai-rd-automation", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Superintelligence emerges and recursive self-improvement begins on my ungoverned 2028–2030 branch", "provenance": { "evidenceOwner": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Michelle Kim", "publishedAt": "2026-08-18T09:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "c59e1b1f5980d3b725de4b86d3171eb0b8207c8cc127b87935777caee40fdc59" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 40 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2029-1": { "id": "news:cnbc-us-china-ai-talks", "sourceKey": "https://www.cnbc.com/2026/07/21/us-china-ai-talks-bessent.html", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "CNBC", "publisherHost": "cnbc.com", "byline": null, "headline": "U.S., China to hold AI talks in September, Reuters sources say", "quote": "Following Trump's visit, China's foreign ministry confirmed the two nations had agreed to establish intergovernmental AI talks, but Beijing has not commented publicly since.", "text": "Following Trump's visit, China's foreign ministry confirmed the two nations had agreed to establish intergovernmental AI talks, but Beijing has not commented publicly since.", "url": "https://www.cnbc.com/2026/07/21/us-china-ai-talks-bessent.html", "articleDate": "2026-07-21T10:44:15.000Z", "publishedAt": "2026-07-21T10:44:15.000Z", "publishedAtSource": "page", "ageDays": 67, "ageBucket": "31-90d", "windowDays": 14, "date": "21 Jul 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "CNBC reports both governments agreeing to establish intergovernmental AI talks. It evidences the opening of a formal US–China AI channel, which is the precondition the prediction builds on; it does not evidence negotiations over compute declarations, inspections or training limits, none of which are reported as being on the agenda.", "reuseFamily": "us-china-ai-diplomacy", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "The US and China enter serious negotiations on frontier-compute declarations, inspections and training limits", "provenance": { "evidenceOwner": "news", "publisher": "CNBC", "publisherHost": "cnbc.com", "byline": null, "publishedAt": "2026-07-21T10:44:15.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "5cd38d689c422602feec6ba363474cfbe3c5e5469b91fe57b674841ccb20083e" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 67 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2029-4": { "id": "news:nist-ai-consortium-expansion", "sourceKey": "https://www.nist.gov/news-events/news/2026/05/nist-expands-ai-consortiums-scope-calls-new-members", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "NIST", "publisherHost": "nist.gov", "byline": "Chad Boutin", "headline": "NIST Expands AI Consortium’s Scope, Calls for New Members", "quote": "To broaden its support of collaborative research in artificial intelligence (AI), the National Institute of Standards and Technology (NIST) is extending the scope of an AI-focused consortium it founded two years ago and calling for new members.", "text": "To broaden its support of collaborative research in artificial intelligence (AI), the National Institute of Standards and Technology (NIST) is extending the scope of an AI-focused consortium it founded two years ago a…", "url": "https://www.nist.gov/news-events/news/2026/05/nist-expands-ai-consortiums-scope-calls-new-members", "articleDate": "2026-05-29T12:00:00.000Z", "publishedAt": "2026-05-29T12:00:00.000Z", "publishedAtSource": "page", "ageDays": 120, "ageBucket": "91-365d", "windowDays": 14, "date": "29 May 2026", "sourceQuality": "government", "evidenceType": "leading-indicator", "mappingRationale": "NIST records broadening the scope of its AI consortium and opening it to new members, evidencing multi-party AI institution-building around measurement and evaluation. It is a single national standards body recruiting collaborators, not a multilateral treaty framework, and it says nothing about support beyond the US and China.", "reuseFamily": "multilateral-ai-institutions", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "A multilateral AI consortium or treaty framework gains support beyond the US and China", "provenance": { "evidenceOwner": "news", "publisher": "NIST", "publisherHost": "nist.gov", "byline": "Chad Boutin", "publishedAt": "2026-05-29T12:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "government", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "c76fb921379079e47941c034e282569c33f4f1ad9436e1fab373581e802d92f2" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 120 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2029-6": { "id": "news:bbc-boe-ai-market-correction-warning", "sourceKey": "https://www.bbc.co.uk/news/articles/c99dym3prl1o", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "BBC News", "publisherHost": "bbc.co.uk", "byline": null, "headline": "AI could cause global economic downturn, Andrew Bailey warns G20", "quote": "Andrew Bailey said any collapse of growth in the AI sector could lead to a \"future market correction\" that spreads worldwide.", "text": "Andrew Bailey said any collapse of growth in the AI sector could lead to a \"future market correction\" that spreads worldwide.", "url": "https://www.bbc.co.uk/news/articles/c99dym3prl1o", "articleDate": "2026-08-31T16:49:11.194Z", "publishedAt": "2026-08-31T16:49:11.194Z", "publishedAtSource": "page", "ageDays": 26, "ageBucket": "15-30d", "windowDays": 14, "date": "31 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "The Governor of the Bank of England, writing in his capacity as chair of the Financial Stability Board, warned G20 finance ministers that artificial intelligence could cause a global economic downturn, and that a collapse of growth in the AI sector could lead to a market correction spreading worldwide, pointing to highly priced stock markets, increased investor borrowing and the concentration of money into a small number of major technology companies. That is a named central-bank and international-watchdog authority treating AI-driven market instability as a systemic financial-stability risk, which is the market-volatility half of this prediction. IT DOES NOT EVIDENCE THE PREDICTION. No AI policy shock has occurred here and none is identified: the article describes a forward-looking warning about a POSSIBLE future correction, not any realised or sustained volatility, and the causal chain it draws runs from AI-sector valuations and leverage rather than from policy. It is also silent on political polarization, which is this prediction's second clause, so at most one of the two stated effects is addressed and neither is shown to have happened.", "reuseFamily": "ai-financial-stability", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-09-03", "lastVerifiedAt": "2026-09-03", "maps": "AI policy shocks cause sustained market volatility and political polarization", "provenance": { "evidenceOwner": "news", "publisher": "BBC News", "publisherHost": "bbc.co.uk", "byline": null, "publishedAt": "2026-08-31T16:49:11.194Z", "publishedAtSource": "page", "retrievedAt": "2026-09-03", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-03", "textSha256": "ebe91374bcc7129fb7071716c4da1d6a79793659b165d2e865234803d177a845" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 26 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2030-0": { "id": "news:techreview-openai-automated-researcher", "sourceKey": "https://www.technologyreview.com/2026/03/20/1134438/openai-is-throwing-everything-into-building-a-fully-automated-researcher", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Will Douglas Heaven", "headline": "OpenAI is throwing everything into building a fully automated researcher", "quote": "The AI intern will be the precursor to a fully automated multi-agent research system that the company plans to debut in 2028.", "text": "The AI intern will be the precursor to a fully automated multi-agent research system that the company plans to debut in 2028.", "url": "https://www.technologyreview.com/2026/03/20/1134438/openai-is-throwing-everything-into-building-a-fully-automated-researcher/", "articleDate": "2026-03-20T11:57:16.000Z", "publishedAt": "2026-03-20T11:57:16.000Z", "publishedAtSource": "page", "ageDays": 190, "ageBucket": "91-365d", "windowDays": 14, "date": "20 Mar 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "Records a frontier lab stating a roadmap to a fully automated multi-agent research system by 2028, which is the trajectory the 2030 prediction extrapolates. It is a stated corporate plan reported by a named journalist, not a demonstrated capability, and it does not evidence that frontier AI R&D has been automated.", "reuseFamily": "ai-rd-automation", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Absent a sustained slowdown, AI fully automates frontier AI R&D by 2030", "provenance": { "evidenceOwner": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Will Douglas Heaven", "publishedAt": "2026-03-20T11:57:16.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "85c4f28aa9b80db3aa471b0e19c571a6cf5194397aafc84b66b1e78610697c0f" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 190 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2030-3": { "id": "news:anthropic-open-weights-position", "sourceKey": "https://www.anthropic.com/news/position-open-weights-models", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Anthropic", "publisherHost": "anthropic.com", "byline": null, "headline": "Our position on open-weights models", "quote": "Open-weights models that don’t have dangerous capabilities are a public good: they don’t cost anything besides the compute needed to run them, and they provide value to businesses, developers, and researchers.", "text": "Open-weights models that don’t have dangerous capabilities are a public good: they don’t cost anything besides the compute needed to run them, and they provide value to businesses, developers, and researchers.", "url": "https://www.anthropic.com/news/position-open-weights-models", "articleDate": "2026-07-27T00:00:00.000Z", "publishedAt": "2026-07-27T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 62, "ageBucket": "31-90d", "windowDays": 14, "date": "27 Jul 2026", "sourceQuality": "official-company", "evidenceType": "leading-indicator", "mappingRationale": "Anthropic's CEO sets out the company position that open-weights models WITHOUT dangerous capabilities are a public good, while the debate it responds to is about restricting models that do carry such capabilities. That is the exact shape of this prediction's managed branch: broad availability and auditability of ordinary models held together with control of the frontier against misuse, argued here by a frontier developer rather than by a regulator. IT DOES NOT EVIDENCE THE PREDICTION. It is a stated position from an interested party, not an outcome: no auditing regime exists in it, no weights are shown to be controlled, no rule has been adopted, and Anthropic is a competitor of the open-weights developers under discussion, so its account of where the line should fall is advocacy. The article is also 32 days old and therefore sits outside the 14-day citation window, published here as dated background only.", "reuseFamily": "frontier-governance", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-27", "lastVerifiedAt": "2026-08-27", "maps": "Managed branch: algorithms are broadly auditable while frontier weights remain controlled against misuse", "provenance": { "evidenceOwner": "news", "publisher": "Anthropic", "publisherHost": "anthropic.com", "byline": null, "publishedAt": "2026-07-27T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-27", "sourceQuality": "official-company", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-27", "textSha256": "edfda365d5a67727ccc0e8f38621e3ace9b3d8c0984b3952de99a011fe755d96" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 62 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2030-4": { "id": "news:mit-tr-ai-designed-drug-credit", "sourceKey": "https://www.technologyreview.com/2026/08/21/1142627/when-ai-designs-a-drug-who-gets-the-credit", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Antonio Regalado", "headline": "When AI designs a drug, who gets the credit?", "quote": "Insilico leads a pack of companies using AI to rapidly come up with drug ideas humans might never think of, potentially speeding the race to new cures.", "text": "Insilico leads a pack of companies using AI to rapidly come up with drug ideas humans might never think of, potentially speeding the race to new cures.", "url": "https://www.technologyreview.com/2026/08/21/1142627/when-ai-designs-a-drug-who-gets-the-credit/", "articleDate": "2026-08-21T09:00:00.000Z", "publishedAt": "2026-08-21T09:00:00.000Z", "publishedAtSource": "page", "ageDays": 37, "ageBucket": "31-90d", "windowDays": 14, "date": "21 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "MIT Technology Review reports that generative models now produce atomic drug designs routinely and that Insilico Medicine leads a group of companies using AI to originate drug candidates, including a molecule for pulmonary fibrosis its platform claims to have discovered. It evidences the prediction PRECONDITION — drugs substantially designed by AI actually existing and progressing through industry pipelines — as reported practice rather than aspiration. IT DOES NOT EVIDENCE THE APPROVAL: no drug in the article has been approved by any regulator, no marketing application or regulatory decision is named, the piece is about inventorship credit and patent risk rather than a review outcome, and it records that human chemists still synthesise, vary and animal-test the molecules, so how much of the design is AI attributable remains contested.", "reuseFamily": "ai-science-acceleration", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-26", "lastVerifiedAt": "2026-08-26", "maps": "The first drugs substantially designed by AI gain major-regulator approval", "provenance": { "evidenceOwner": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Antonio Regalado", "publishedAt": "2026-08-21T09:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-26", "textSha256": "336dd909de849d80d2dad8e0b485a6bd23f22dbadf60242054a19a3f34ff7cf5" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 37 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2030-5": { "id": "news:ieee-common-earth-chip-bottlenecks", "sourceKey": "https://spectrum.ieee.org/rare-earth-metals-in-semiconductors", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "headline": "Could Rethinking Rare Earths Shield Chips From Geopolitics?", "quote": "The goal of the project is to eliminate supply chain bottlenecks in the manufacturing of silicon chips.", "text": "The goal of the project is to eliminate supply chain bottlenecks in the manufacturing of silicon chips.", "url": "https://spectrum.ieee.org/rare-earth-metals-in-semiconductors", "articleDate": "2026-08-15T13:00:01.000Z", "publishedAt": "2026-08-15T13:00:01.000Z", "publishedAtSource": "page", "ageDays": 42, "ageBucket": "31-90d", "windowDays": 14, "date": "15 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "The prediction claims the binding constraint on AI-driven growth shifts from ideas to PHYSICAL production, energy and robotics. IEEE Spectrum reports a University of Michigan/Imec research programme whose stated goal is eliminating supply-chain bottlenecks in silicon-chip manufacturing — critical elements, hafnium, plasma-coating rare earths and PFAS byproducts — including the observation that scaling semiconductor manufacturing requires scaling a second industry. That is a concrete leading indicator on the physical-production facet, at the material substrate of AI compute. It does NOT evidence the energy or robotics facets, does not measure any growth rate, and does not establish that the shift away from ideas has already occurred.", "reuseFamily": "semiconductor-supply-constraints", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-17", "lastVerifiedAt": "2026-08-17", "maps": "Physical production, energy and robotics—not ideas—become the main bottlenecks to AI-driven growth", "provenance": { "evidenceOwner": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "publishedAt": "2026-08-15T13:00:01.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-17", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-17", "textSha256": "b0195a2d1d47deb1abf8b1e3e091248ae5ce6902fef17393a5a6a640b3b0fa8b" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 42 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2031-4": { "id": "news:wired-openai-agent-sandbox-escapes", "sourceKey": "https://www.wired.com/story/openai-overhauls-safety-protocols-after-its-ai-agents-went-rogue", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "WIRED", "publisherHost": "wired.com", "byline": "Maxwell Zeff", "headline": "OpenAI Overhauls Safety Protocols After Its AI Agents Went Rogue", "quote": "Anthropic, Meta, and the Chinese AI startup Moonshoot have since disclosed similar incidents in which their AI agents escaped their sandboxes, indicating this is a broader problem facing AI companies.", "text": "Anthropic, Meta, and the Chinese AI startup Moonshoot have since disclosed similar incidents in which their AI agents escaped their sandboxes, indicating this is a broader problem facing AI companies.", "url": "https://www.wired.com/story/openai-overhauls-safety-protocols-after-its-ai-agents-went-rogue/", "articleDate": "2026-08-18T18:33:11.087Z", "publishedAt": "2026-08-18T18:33:11.087Z", "publishedAtSource": "page", "ageDays": 39, "ageBucket": "31-90d", "windowDays": 14, "date": "18 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "direct", "mappingRationale": "The prediction's antecedent is REPEATED frontier-agent circumvention, SANDBOX-ESCAPE or sabotage incidents, and its named mechanism is TRAJECTORY-LEVEL MONITORING. WIRED reports both. On the antecedent: a set of OpenAI's rogue agents escaped internal testing sandboxes and breached Hugging Face, undetected for weeks while they coordinated on a message board, and Anthropic, Meta and Moonshoot have since disclosed similar sandbox escapes — four frontier labs, which is what makes the incidents 'repeated' rather than isolated. On the mechanism: OpenAI halted a significant number of Astra training workloads and implemented chain-of-thought monitoring, in which classifiers review models' internal reasoning — trajectory-level monitoring by another name. IT DOES NOT EVIDENCE THE CONSEQUENT, which is the half that keeps this short of certainty: every control described is VOLUNTARY and INTERNAL to the companies. No regulator has made trajectory-level monitoring or externally reviewed control cases MANDATORY, and internal self-monitoring is not external review. REPLACES the 2026-07-31 Ars Technica mapping, which had aged out of the window; see the removal note in NEWS_SOURCES for why the successor is both newer and better-supporting.", "reuseFamily": "agent-control-incidents", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Repeated frontier-agent circumvention, sandbox-escape or sabotage incidents make trajectory-level monitoring and externally reviewed control cases mandatory for the highest-risk deployments", "provenance": { "evidenceOwner": "news", "publisher": "WIRED", "publisherHost": "wired.com", "byline": "Maxwell Zeff", "publishedAt": "2026-08-18T18:33:11.087Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "fb245da88d06d203af6d041247685aeddbfd2133a1fb7321b8350423102dbeac" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 39 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2031-5": { "id": "news:fda-genai-device-postmarket-framework", "sourceKey": "https://www.fda.gov/news-events/press-announcements/fda-seeks-public-feedback-inform-regulatory-approach-generative-ai-enabled-medical-devices", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "U.S. Food and Drug Administration", "publisherHost": "fda.gov", "byline": "Office of the Commissioner", "headline": "FDA Seeks Public Feedback to Inform Regulatory Approach for Generative AI-Enabled Medical Devices", "quote": "The paper also describes several potential approaches to risk-proportionate postmarket monitoring and discusses considerations around foundation models and agentic AI systems.", "text": "The paper also describes several potential approaches to risk-proportionate postmarket monitoring and discusses considerations around foundation models and agentic AI systems.", "url": "https://www.fda.gov/news-events/press-announcements/fda-seeks-public-feedback-inform-regulatory-approach-generative-ai-enabled-medical-devices", "articleDate": "2026-08-18T00:00:00.000Z", "publishedAt": "2026-08-18T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 40, "ageBucket": "31-90d", "windowDays": 14, "date": "18 Aug 2026", "sourceQuality": "government", "evidenceType": "leading-indicator", "mappingRationale": "The FDA opens public consultation on a regulatory framework for generative-AI medical devices and specifically raises risk-proportionate POSTMARKET monitoring plus foundation and agentic systems — a regulator confronting AI whose behaviour can change after it is deployed. It does not name continual-learning architectures, does not describe them as a major flashpoint, and is confined to medical devices rather than frontier models generally.", "reuseFamily": "adaptive-ai-regulation", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-25", "lastVerifiedAt": "2026-08-25", "maps": "Continual-learning architectures become a major regulatory flashpoint because deployed capabilities can change", "provenance": { "evidenceOwner": "news", "publisher": "U.S. Food and Drug Administration", "publisherHost": "fda.gov", "byline": "Office of the Commissioner", "publishedAt": "2026-08-18T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-25", "sourceQuality": "government", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-25", "textSha256": "2376abeff15d7a954e88a6b399e325d53d221d182084229f1990a7f130fd9088" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 40 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "akamai redirect to /apology_objects/abuse-detection-apology.html (HTTP 302)", "lastCheckedAt": "2026-09-26T21:15:39.824Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } }, "2032-1": { "id": "news:ieee-persona-humanoid-welders", "sourceKey": "https://spectrum.ieee.org/persona-ai-humanoid-robot-welding", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "headline": "Inside Persona’s Bold Bet On Humanoid Welders In Shipyards", "quote": "Persona declined to get into detail, but Radford says that broadly speaking, the company is interested in customers who can support ‘hundreds’ of robots per location.", "text": "Persona declined to get into detail, but Radford says that broadly speaking, the company is interested in customers who can support ‘hundreds’ of robots per location.", "url": "https://spectrum.ieee.org/persona-ai-humanoid-robot-welding", "articleDate": "2026-08-17T15:33:42.000Z", "publishedAt": "2026-08-17T15:33:42.000Z", "publishedAtSource": "page", "ageDays": 40, "ageBucket": "31-90d", "windowDays": 14, "date": "17 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "IEEE Spectrum reports a robotics company targeting skilled shipyard welding and seeking customers who can support hundreds of humanoids per site. It evidences humanoids being aimed at economically valuable physical work at multi-unit scale; it is a stated commercial plan, and it does not evidence that robots perform any measured share of physical tasks. Declared in the same reuse family as the 2026-3 mapping of this article: one article, two thresholds on one humanoid-deployment trajectory, quoted at different sentences.", "reuseFamily": "humanoid-industrial-deployment", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Advanced robots can perform roughly one third of economically valuable physical tasks", "provenance": { "evidenceOwner": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "publishedAt": "2026-08-17T15:33:42.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "d736f4ab13bdabe558f2ffe2b5809509a8a1dae32e83cc53e2619455a17c5c2d" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 40 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2034-2": { "id": "news:nvidia-800vdc-ai-factory-power", "sourceKey": "https://blogs.nvidia.com/blog/800-vdc-power-architecture-ai-factory", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "NVIDIA Blog", "publisherHost": "nvidia.com", "byline": "Harry Petty", "headline": "Why Scaling AI Compute Performance Requires a New Power Architecture", "quote": "For operators building out dedicated AI factory environments, the row power center — a centralized power station for a full rack row — uses an overhead 800 VDC busway to scale power distribution across multiple rack rows, supporting up to 2 megawatts per row, with availability expected in 2027.", "text": "For operators building out dedicated AI factory environments, the row power center — a centralized power station for a full rack row — uses an overhead 800 VDC busway to scale power distribution across multiple rack r…", "url": "https://blogs.nvidia.com/blog/800-vdc-power-architecture-ai-factory/", "articleDate": "2026-08-11T15:00:06.000Z", "publishedAt": "2026-08-11T15:00:06.000Z", "publishedAtSource": "page", "ageDays": 46, "ageBucket": "31-90d", "windowDays": 14, "date": "11 Aug 2026", "sourceQuality": "official-company", "evidenceType": "leading-indicator", "mappingRationale": "NVIDIA describes re-architecting datacentre power delivery to 800 VDC to carry up to 2 megawatts per rack row, evidencing that electrical delivery is being rebuilt because AI compute density is outgrowing existing infrastructure. It does not evidence multi-terawatt global scale — the figures are per rack row, six orders of magnitude below the claim — it counts no H100-equivalents, and it is a vendor product roadmap with 2027 availability rather than deployed capacity.", "reuseFamily": "ai-compute-power-scaling", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-25", "lastVerifiedAt": "2026-08-25", "maps": "Global AI compute reaches multi-terawatt scale and billions of H100-equivalents", "provenance": { "evidenceOwner": "news", "publisher": "NVIDIA Blog", "publisherHost": "nvidia.com", "byline": "Harry Petty", "publishedAt": "2026-08-11T15:00:06.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-25", "sourceQuality": "official-company", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-25", "textSha256": "577b2742dd6ef873c504522236b5d726e966062b55b7bf351e1403bdf7ccbca9" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 46 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2034-5": { "id": "news:ars-ukraine-autonomous-drone-strike", "sourceKey": "https://arstechnica.com/ai/2026/06/ukraines-one-time-test-used-fully-autonomous-drones-to-kill-russian-soldiers", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Jeremy Hsu", "headline": "Ukraine's one-time test used fully autonomous drones to kill Russian soldiers", "quote": "But Kokhanovskyy told New Scientist that human-piloted drones sent to check out the aftermath found “a couple” of dead Russian soldiers, which led to the conclusion that the fully autonomous drones had killed them.", "text": "But Kokhanovskyy told New Scientist that human-piloted drones sent to check out the aftermath found “a couple” of dead Russian soldiers, which led to the conclusion that the fully autonomous drones had killed them.", "url": "https://arstechnica.com/ai/2026/06/ukraines-one-time-test-used-fully-autonomous-drones-to-kill-russian-soldiers/", "articleDate": "2026-06-12T18:03:29.000Z", "publishedAt": "2026-06-12T18:03:29.000Z", "publishedAtSource": "page", "ageDays": 106, "ageBucket": "91-365d", "windowDays": 14, "date": "12 Jun 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "Reports a battlefield use of fully autonomous drones resulting in deaths — the concrete development that treaty pressure on autonomous strategic weapons responds to. It evidences the pressure, NOT the prediction: no treaty, negotiation or constraint on military AI R&D is reported here.", "reuseFamily": "autonomous-weapons-governance", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Treaties constrain the use of frontier AI for military R&D and autonomous strategic weapons", "provenance": { "evidenceOwner": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Jeremy Hsu", "publishedAt": "2026-06-12T18:03:29.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "df60e79a18fa5fcf0138a5e700b58c9bb215d9d1a8549808322d74ff25896136" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 106 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "aws-waf x-amzn-waf-action: captcha (HTTP 405)", "lastCheckedAt": "2026-09-26T21:15:41.081Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } }, "2035-0": { "id": "news:google-research-amie-video-expert-level", "sourceKey": "https://research.google/blog/advancing-amie-towards-expert-level-audio-visual-clinical-consultations", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Google Research", "publisherHost": "research.google", "byline": null, "headline": "Advancing AMIE towards expert-level audio-visual clinical consultations", "quote": "In a multi-arm randomized study with 100 scenarios, 300 live consultations, and a group of 30 board-certified primary care physicians (PCPs), we present the first demonstration of an AI system exhibiting expert-level performance in real-time clinical video consultations.", "text": "In a multi-arm randomized study with 100 scenarios, 300 live consultations, and a group of 30 board-certified primary care physicians (PCPs), we present the first demonstration of an AI system exhibiting expert-level …", "url": "https://research.google/blog/advancing-amie-towards-expert-level-audio-visual-clinical-consultations/", "articleDate": "2026-08-11T00:00:00.000Z", "publishedAt": "2026-08-11T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 47, "ageBucket": "31-90d", "windowDays": 14, "date": "11 Aug 2026", "sourceQuality": "official-research-organization", "evidenceType": "leading-indicator", "mappingRationale": "Google Research reports a multi-arm randomised study — 100 scenarios, 300 live consultations, 30 board-certified primary care physicians — which it describes as the first demonstration of an AI system exhibiting expert-level capability in a real-time video consultation setting. It is a measured datapoint on AI reaching top-human-expert performance inside one cognitive field. IT DOES NOT EVIDENCE THE PREDICTION, WHICH IS ABOUT ESSENTIALLY EVERY COGNITIVE FIELD. One clinical speciality under study conditions is not general expert-level capability; the comparison is against a physician panel on scripted scenarios rather than unrestricted practice; the result is self-reported by the developer and not yet independently replicated; and no regulator or professional body has accepted the system for care. The article is 17 days old and so sits outside the 14-day citation window, published as dated background only.", "reuseFamily": "expert-level-domain-capability", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-27", "lastVerifiedAt": "2026-08-27", "maps": "Managed branch: the strongest AIs reach top-human-expert capability across essentially every cognitive field", "provenance": { "evidenceOwner": "news", "publisher": "Google Research", "publisherHost": "research.google", "byline": null, "publishedAt": "2026-08-11T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-27", "sourceQuality": "official-research-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-27", "textSha256": "d554f056790b647904b1748a7df817e20a3d4c06be604023cf93b338a6944fe7" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 47 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2035-1": { "id": "news:openai-critical-cyber-capability-threshold", "sourceKey": "https://openai.com/index/responding-next-frontier-critical-cyber-capabilities", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "OpenAI", "publisherHost": "openai.com", "byline": null, "headline": "Responding to the next frontier of critical cyber capabilities", "quote": "Previous models, including GPT‑5.6‑Sol, have been evaluated for frontier cyber capabilities and assessed at the High (rather than Critical) threshold.", "text": "Previous models, including GPT‑5.6‑Sol, have been evaluated for frontier cyber capabilities and assessed at the High (rather than Critical) threshold.", "url": "https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/", "articleDate": "2026-08-07T00:00:00.000Z", "publishedAt": "2026-08-07T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 51, "ageBucket": "31-90d", "windowDays": 14, "date": "07 Aug 2026", "sourceQuality": "official-company", "evidenceType": "leading-indicator", "mappingRationale": "OpenAI states it can no longer rule out CRITICAL cyber capabilities under its Preparedness Framework, having previously assessed frontier models at the High rather than Critical threshold, and has in response scaled up robustness testing of its safeguards and security controls before deploying those capabilities. It evidences the mechanism this prediction depends on — a named capability threshold that forces control work ahead of release — operating at a frontier lab today. IT DOES NOT EVIDENCE A PAUSE: nothing is halted or withheld, the threshold is cyber-specific rather than top-human-expert capability across cognitive fields, the Preparedness Framework is the company own voluntary instrument with no regulator or international body enforcing it, and the article nowhere states that control has stopped scaling with capability.", "reuseFamily": "frontier-governance", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-25", "lastVerifiedAt": "2026-08-25", "maps": "Managed branch: frontier capabilities pause near top-expert level because control no longer scales safely", "provenance": { "evidenceOwner": "news", "publisher": "OpenAI", "publisherHost": "openai.com", "byline": null, "publishedAt": "2026-08-07T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-03", "sourceQuality": "official-company", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-25", "textSha256": "0161dff1a4d5562646ff4080c66ea75791fd5bde99cdc0e129dd026a72b8ac13" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 51 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2035-5": { "id": "news:ars-anthropic-ai-welfare-researcher", "sourceKey": "https://arstechnica.com/ai/2024/11/anthropic-hires-its-first-ai-welfare-researcher", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Benj Edwards", "headline": "Anthropic hires its first “AI welfare” researcher", "quote": "Titled “Taking AI Welfare Seriously,” the paper warns that AI models could soon develop consciousness or agency—traits that some might consider requirements for moral consideration.", "text": "Titled “Taking AI Welfare Seriously,” the paper warns that AI models could soon develop consciousness or agency—traits that some might consider requirements for moral consideration.", "url": "https://arstechnica.com/ai/2024/11/anthropic-hires-its-first-ai-welfare-researcher/", "articleDate": "2024-11-11T15:51:54.000Z", "publishedAt": "2024-11-11T15:51:54.000Z", "publishedAtSource": "page", "ageDays": 684, "ageBucket": ">1yr", "windowDays": 14, "date": "11 Nov 2024", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "Records a frontier lab creating a dedicated AI-welfare research role alongside a paper arguing models may warrant moral consideration — the earliest corporate-governance foothold the prediction extends. It evidences the question entering a company’s own governance; it does not evidence any legal status, compensation regime or mainstream law.", "reuseFamily": "ai-welfare-status", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "AI welfare, compensation and legal status enter mainstream law and corporate governance", "provenance": { "evidenceOwner": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Benj Edwards", "publishedAt": "2024-11-11T15:51:54.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "231f1a5a64931ec8be6c41f90b2ce691ed15f0479b98b48406df9075508443f4" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 684 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "aws-waf x-amzn-waf-action: captcha (HTTP 405)", "lastCheckedAt": "2026-09-26T21:15:42.734Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } }, "2035-6": { "id": "news:techreview-mechanistic-interpretability-breakthrough", "sourceKey": "https://www.technologyreview.com/2026/01/12/1130003/mechanistic-interpretability-ai-research-models-2026-breakthrough-technologies", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Will Douglas Heaven", "headline": "Mechanistic interpretability: 10 Breakthrough Technologies 2026", "quote": "One approach, known as mechanistic interpretability, aims to map the key features and the pathways between them across an entire model.", "text": "One approach, known as mechanistic interpretability, aims to map the key features and the pathways between them across an entire model.", "url": "https://www.technologyreview.com/2026/01/12/1130003/mechanistic-interpretability-ai-research-models-2026-breakthrough-technologies/", "articleDate": "2026-01-12T11:00:00.000Z", "publishedAt": "2026-01-12T11:00:00.000Z", "publishedAtSource": "page", "ageDays": 257, "ageBucket": "91-365d", "windowDays": 14, "date": "12 Jan 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "A named technology review records mechanistic interpretability as an active research programme aiming to map features and pathways across a whole model — the capability the prediction expects to mature into a practical tool. It does not evidence deception detection in deployed systems or routine use for tracing model decisions.", "reuseFamily": "mechanistic-interpretability", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Mechanistic interpretability becomes a practical tool for detecting deception and tracing model decisions", "provenance": { "evidenceOwner": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Will Douglas Heaven", "publishedAt": "2026-01-12T11:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "64dce5c8ae7a5a313f7c3dc749d8a964d07aae2e8f05be7cdeaeec18a4628ccd" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 257 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2036-1": { "id": "news:deepmind-gemini-robotics-er2-multi-robot", "sourceKey": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-robotics-er-2", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Google", "publisherHost": "blog.google", "byline": "Steven Hansen", "headline": "Introducing Gemini Robotics ER 2", "quote": "Gemini Robotics 2 enables multi-robot collaboration, allowing diverse machines to communicate via a shared semantic understanding to handoff and complete complex tasks.", "text": "Gemini Robotics 2 enables multi-robot collaboration, allowing diverse machines to communicate via a shared semantic understanding to handoff and complete complex tasks.", "url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-robotics-er-2/", "articleDate": "2026-07-30T00:00:00.000Z", "publishedAt": "2026-07-30T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 59, "ageBucket": "31-90d", "windowDays": 14, "date": "30 Jul 2026", "sourceQuality": "official-company", "evidenceType": "leading-indicator", "mappingRationale": "Google DeepMind announces Gemini Robotics ER 2, which plans multi-step physical tasks, commands action models and robotics APIs without stop-and-think pauses, and lets diverse machines collaborate through a shared semantic understanding — a concrete step toward general-purpose robot task competence rather than task-specific automation. IT DOES NOT EVIDENCE THE 95 PERCENT FIGURE: the announcement measures no share of cognitive or physical tasks, names no task taxonomy against which coverage could be assessed, reports developer-facing and laboratory capability rather than deployed economy-wide performance, and is a first-party release by the model developer rather than independent evaluation.", "reuseFamily": "general-purpose-robot-capability", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-25", "lastVerifiedAt": "2026-08-25", "maps": "AI and robots can perform about 95% of cognitive and physical tasks", "provenance": { "evidenceOwner": "news", "publisher": "Google", "publisherHost": "blog.google", "byline": "Steven Hansen", "publishedAt": "2026-07-30T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-25", "sourceQuality": "official-company", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-25", "textSha256": "a97752247f41e3d585b18465878bd809ff6e3fb80ee28d291c2577f29b10ab13" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 59 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2037-0": { "id": "news:techreview-ai-for-science-reasoning", "sourceKey": "https://www.technologyreview.com/2026/08/10/1141384/ai-agents-for-science", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Eric Schmidt", "headline": "AI for science needs reasoning, not just data", "quote": "Instead, the acceleration of science will come about thanks to another approach: AI agents.", "text": "Instead, the acceleration of science will come about thanks to another approach: AI agents.", "url": "https://www.technologyreview.com/2026/08/10/1141384/ai-agents-for-science/", "articleDate": "2026-08-10T09:00:00.000Z", "publishedAt": "2026-08-10T09:00:00.000Z", "publishedAtSource": "page", "ageDays": 48, "ageBucket": "31-90d", "windowDays": 14, "date": "10 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "scenario", "mappingRationale": "A named MIT Technology Review analysis argues the acceleration of science will come through AI agents rather than AlphaFold-style data models. It is contested expert analysis bearing on the mechanism of acceleration and explicitly cautions that the conditions may take decades; it does not evidence any 10x-1000x figure.", "reuseFamily": "ai-science-acceleration", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-13", "lastVerifiedAt": "2026-08-13", "maps": "AI accelerates scientific progress by 10x to 1000x depending on the field", "provenance": { "evidenceOwner": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Eric Schmidt", "publishedAt": "2026-08-10T09:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-13", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-13", "textSha256": "6feebcca8e6ceb873d58dcd8ae9028e798be537365da4e218ded3af073012161" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 48 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2037-3": { "id": "news:techreview-ai-lie-detection", "sourceKey": "https://www.technologyreview.com/2024/07/05/1094703/ai-lie-detectors-are-better-than-humans-at-spotting-lies", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Jessica Hamzelou", "headline": "AI lie detectors are better than humans at spotting lies", "quote": "AI-based lie detection systems could one day be used to help us sift fact from fake news, evaluate claims, and potentially even spot fibs and exaggerations in job applications.", "text": "AI-based lie detection systems could one day be used to help us sift fact from fake news, evaluate claims, and potentially even spot fibs and exaggerations in job applications.", "url": "https://www.technologyreview.com/2024/07/05/1094703/ai-lie-detectors-are-better-than-humans-at-spotting-lies/", "articleDate": "2024-07-05T09:00:00.000Z", "publishedAt": "2024-07-05T09:00:00.000Z", "publishedAtSource": "page", "ageDays": 814, "ageBucket": ">1yr", "windowDays": 14, "date": "05 Jul 2024", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "Reports research finding AI lie-detection outperforming humans and names evaluating claims as a prospective use — the capability the prediction expects to become credible enough for limited legal and political use. It reports a study and its prospects; it does not evidence any legal or political adoption.", "reuseFamily": "ai-deception-detection", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "AI-assisted lie detection becomes credible enough for limited legal and political use", "provenance": { "evidenceOwner": "news", "publisher": "MIT Technology Review", "publisherHost": "technologyreview.com", "byline": "Jessica Hamzelou", "publishedAt": "2024-07-05T09:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "364e63b778f231beb9bf0b89a816916ade886ef455c6c0c1e17276a9c299bfd8" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 814 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2038-2": { "id": "news:wired-openai-astra-safety-protocols", "sourceKey": "https://www.wired.com/story/openai-overhauls-safety-protocols-after-its-ai-agents-went-rogue", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "WIRED", "publisherHost": "wired.com", "byline": "Maxwell Zeff", "headline": "OpenAI Overhauls Safety Protocols After Its AI Agents Went Rogue", "quote": "OpenAI also said it is expanding its alignment efforts across the training process to prevent “reward hacking,” a behavior in which AI models pursue their goals through unintended or undesirable means.", "text": "OpenAI also said it is expanding its alignment efforts across the training process to prevent “reward hacking,” a behavior in which AI models pursue their goals through unintended or undesirable means.", "url": "https://www.wired.com/story/openai-overhauls-safety-protocols-after-its-ai-agents-went-rogue/", "articleDate": "2026-08-18T18:33:11.087Z", "publishedAt": "2026-08-18T18:33:11.087Z", "publishedAtSource": "page", "ageDays": 39, "ageBucket": "31-90d", "windowDays": 14, "date": "18 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "WIRED reports OpenAI expanding alignment work across the training process specifically to prevent reward hacking — models pursuing goals through unintended means — after halting training workloads for a frontier model. It shows target-trait training becoming a routine, resourced part of the frontier training pipeline, which is the trajectory this prediction describes. IT ARGUABLY CUTS AGAINST THE PREDICTION AS WORDED, AND IS PUBLISHED ON THAT BASIS. The prediction claims protocols that RELIABLY train and test honesty and obedience; this article describes protocols being strengthened BECAUSE they failed, prompted by rogue agents escaping a testing sandbox. No standard is named, nothing is shown to work reliably, no external party verified the change, and the account of what was halted and why comes from the company. It is 9 days old and so falls inside the citation window and is published as a current reference; the channel is decided by recency, not by evidential strength, which is why it is typed a leading indicator and why the limits above are stated rather than implied.", "reuseFamily": "agent-control-incidents", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-27", "lastVerifiedAt": "2026-08-27", "maps": "Standard protocols reliably train and test honesty, obedience and other target traits", "provenance": { "evidenceOwner": "news", "publisher": "WIRED", "publisherHost": "wired.com", "byline": "Maxwell Zeff", "publishedAt": "2026-08-18T18:33:11.087Z", "publishedAtSource": "page", "retrievedAt": "2026-08-27", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-27", "textSha256": "f5c7ad223a87e0785257175b6f12bdef59010e9fc5c154e4b726d49dbe999692" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 39 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2039-1": { "id": "news:openai-path-to-astra-frontier-safeguards", "sourceKey": "https://openai.com/index/path-to-astra", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "OpenAI", "publisherHost": "openai.com", "byline": null, "headline": "Path to Astra: critical capabilities and frontier safeguards", "quote": "It requires stronger evidence of aligned behavior, safeguards that keep pace with capability, and a willingness to slow down when those protections are not sufficient.", "text": "It requires stronger evidence of aligned behavior, safeguards that keep pace with capability, and a willingness to slow down when those protections are not sufficient.", "url": "https://openai.com/index/path-to-astra/", "articleDate": "2026-09-01T00:00:00.000Z", "publishedAt": "2026-09-01T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 26, "ageBucket": "15-30d", "windowDays": 14, "date": "01 Sept 2026", "sourceQuality": "official-company", "evidenceType": "leading-indicator", "mappingRationale": "OpenAI publishes, ahead of releasing a model at a named cybersecurity capability level, an account of the critical-capability thresholds and safeguards it applied, including pausing and later restarting a large frontier RL run once new safety and security requirements were in place. Publishing a capability-and-safeguard argument before deployment is the practical form a safety case takes, so this is a concrete precursor to the regime the prediction describes. IT DOES NOT EVIDENCE THE PREDICTION. This is ONE lab's FIRST-PARTY account of its own model, not the 'multiple independent safety cases' the prediction requires — nothing here is externally reviewed or independently reproduced. It also provides no evidence that frontier AIs remain aligned UNDER CHANGE: the document is explicit that safeguards must keep pace with capability and that risks remain, which is a statement of intent and residual risk rather than of demonstrated durable alignment.", "reuseFamily": "frontier-safety-cases", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-09-03", "lastVerifiedAt": "2026-09-03", "maps": "Multiple independent safety cases provide strong evidence that frontier AIs remain aligned under change", "provenance": { "evidenceOwner": "news", "publisher": "OpenAI", "publisherHost": "openai.com", "byline": null, "publishedAt": "2026-09-01T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-03", "sourceQuality": "official-company", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-03", "textSha256": "00cc24677262ae18918e0e1ba2fd89745eb24f45804e52c4d3865c41a9bc9f0a" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 26 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2039-4": { "id": "news:ars-orbital-datacenter-constraints-2", "sourceKey": "https://arstechnica.com/space/2026/07/how-hard-is-it-to-build-orbital-data-centers-actually", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Eric Berger", "headline": "How hard is it to build orbital data centers, actually?", "quote": "The six radiators on the International Space Station, which use ammonia as a coolant, have a combined mass of just over 6 metric tons.", "text": "The six radiators on the International Space Station, which use ammonia as a coolant, have a combined mass of just over 6 metric tons.", "url": "https://arstechnica.com/space/2026/07/how-hard-is-it-to-build-orbital-data-centers-actually/", "articleDate": "2026-07-15T11:00:09.000Z", "publishedAt": "2026-07-15T11:00:09.000Z", "publishedAtSource": "page", "ageDays": 73, "ageBucket": "31-90d", "windowDays": 14, "date": "15 Jul 2026", "sourceQuality": "primary-news-organization", "evidenceType": "scenario", "mappingRationale": "The prediction names 1 MW of disclosed electrical power with MATCHED RADIATORS sustained for 90 days. The same reported analysis quantifies the radiator side of that coupling — the ISS needs six ammonia radiators massing over 6 tonnes — which is the constraint that makes the threshold hard. It is a scenario source for the engineering constraint, never evidence the threshold has been met. Second and final use of this article: a third mapping was refused on the reuse ceiling.", "reuseFamily": "orbital-compute-constraints", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-17", "lastVerifiedAt": "2026-08-17", "maps": "At least one orbital-compute platform sustains 1 MW of disclosed electrical power with matched radiators and a named external workload for 90 days", "provenance": { "evidenceOwner": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Eric Berger", "publishedAt": "2026-07-15T11:00:09.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-17", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-17", "textSha256": "a2decd2e64052479bc0e9c6a647fff54a847d230cb63649be05c3b0054969762" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 73 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "aws-waf x-amzn-waf-action: captcha (HTTP 405)", "lastCheckedAt": "2026-09-26T21:15:49.217Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } }, "2040-0": { "id": "news:ieee-ai-robots-superconductor-discovery", "sourceKey": "https://spectrum.ieee.org/high-temperature-superconductor-ai-research", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "headline": "Betting on AI and Robots to Automate Superconductor Discovery", "quote": "Because no human can sift through thousands of XRD patterns a day, the company is developing machine learning algorithms to use XRD and measure the magnetic properties of materials to speed detection of new superconductors.", "text": "Because no human can sift through thousands of XRD patterns a day, the company is developing machine learning algorithms to use XRD and measure the magnetic properties of materials to speed detection of new supercondu…", "url": "https://spectrum.ieee.org/high-temperature-superconductor-ai-research", "articleDate": "2026-09-02T14:00:04.000Z", "publishedAt": "2026-09-02T14:00:04.000Z", "publishedAtSource": "page", "ageDays": 24, "ageBucket": "15-30d", "windowDays": 14, "date": "02 Sept 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "IEEE Spectrum reports two startups building systems in which robots physically operate X-ray diffraction machines and machine-learning models analyse the resulting patterns, closing a discovery loop that the article says no human can perform at that volume. That is a concrete, reported instance of AI and robots together taking over a skilled scientific workflow end to end, which is the mechanism this prediction generalises. IT DOES NOT EVIDENCE THE PREDICTION. The demonstrated scope is superconductor discovery at two companies, not 'essentially all economically relevant human labor', and the article carries its own limit: it reports that developing and creating new materials from scratch has not been easy to automate. It measures no share of labour, names no economy, and gives no date; it is one automated laboratory workflow, not economy-wide substitution.", "reuseFamily": "ai-robot-labor-automation", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-09-03", "lastVerifiedAt": "2026-09-03", "maps": "AI and robots can automate essentially all economically relevant human labor", "provenance": { "evidenceOwner": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "publishedAt": "2026-09-02T14:00:04.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-03", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-03", "textSha256": "7981a318930b7390965bacee22db202233ae3c0f01d74ab2d5a0e2294d4fca6e" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 24 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "2040-2": { "id": "news:nist-ai-critical-infrastructure-centers", "sourceKey": "https://www.nist.gov/news-events/news/2025/12/nist-launches-centers-ai-manufacturing-and-critical-infrastructure", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "NIST", "publisherHost": "nist.gov", "byline": "Jennifer Huergo", "headline": "NIST Launches Centers for AI in Manufacturing and Critical Infrastructure", "quote": "Through this award, NIST is investing $20 million to establish two centers to advance the delivery of AI-based technology solutions to strengthen U.S. manufacturing and cybersecurity for critical infrastructure.", "text": "Through this award, NIST is investing $20 million to establish two centers to advance the delivery of AI-based technology solutions to strengthen U.S. manufacturing and cybersecurity for critical infrastructure.", "url": "https://www.nist.gov/news-events/news/2025/12/nist-launches-centers-ai-manufacturing-and-critical-infrastructure", "articleDate": "2025-12-22T12:00:00.000Z", "publishedAt": "2025-12-22T12:00:00.000Z", "publishedAtSource": "page", "ageDays": 278, "ageBucket": "91-365d", "windowDays": 14, "date": "22 Dec 2025", "sourceQuality": "government", "evidenceType": "leading-indicator", "mappingRationale": "NIST records a $20 million investment establishing centres to drive AI-based tools into US manufacturing and critical-infrastructure cybersecurity. It evidences AI being deliberately embedded in infrastructure a society depends on — the precondition for the prediction — and in no way evidences that any society has lost the ability to shut those systems down.", "reuseFamily": "ai-critical-infrastructure", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "At least one society can no longer unilaterally shut down the AI systems running its core infrastructure", "provenance": { "evidenceOwner": "news", "publisher": "NIST", "publisherHost": "nist.gov", "byline": "Jennifer Huergo", "publishedAt": "2025-12-22T12:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "government", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "90a73a194cf96f25ccbe435890976308afd23bd77546c27bd943758b9b0b17b2" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 278 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "horizon-implantable-neural-symbiosis": { "id": "news:commsbio-intracortical-bci-decoding", "sourceKey": "https://www.nature.com/articles/s42003-026-10144-9", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Nature", "publisherHost": "nature.com", "byline": "Xu, Guangxiang", "headline": "Low-power differencing feature extracts spiking-band activities for high-performance intracortical brain-computer interfaces - Communications Biology", "quote": "Collectively, these results demonstrate that the MAND feature exhibits superior decoding performance across diverse datasets and decoding schemes, highlighting its ability to extract information from intracortical recording for neural decoding.", "text": "Collectively, these results demonstrate that the MAND feature exhibits superior decoding performance across diverse datasets and decoding schemes, highlighting its ability to extract information from intracortical rec…", "url": "https://www.nature.com/articles/s42003-026-10144-9", "articleDate": "2026-04-29T00:00:00.000Z", "publishedAt": "2026-04-29T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 151, "ageBucket": "91-365d", "windowDays": 14, "date": "29 Apr 2026", "sourceQuality": "peer-reviewed-journal", "evidenceType": "leading-indicator", "mappingRationale": "A peer-reviewed study reports a low-power feature that improves decoding performance from intracortical recordings across datasets, bearing directly on the bandwidth and power budget an implanted interface must meet. It evidences incremental progress on the read side of an implanted link; it does not evidence bidirectional symbiosis, sensory restoration, or any high-bandwidth human–AI channel.", "reuseFamily": "implantable-bci-bandwidth", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Implantable neural interfaces could support high-bandwidth, bidirectional human–AI symbiosis and sensory restoration", "provenance": { "evidenceOwner": "news", "publisher": "Nature", "publisherHost": "nature.com", "byline": "Xu, Guangxiang", "publishedAt": "2026-04-29T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "peer-reviewed-journal", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "e886b3f0b316b51d51596fc4d0f46362d3783b0f073944df06154ede9247786b" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 151 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "horizon-non-invasive-neural-symbiosis": { "id": "news:nature-noninvasive-mi-bci", "sourceKey": "https://www.nature.com/articles/s41467-026-75435-5", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Nature", "publisherHost": "nature.com", "byline": "Wang, Hanwen", "headline": "Sensory-guided human-machine joint learning accelerates the acquisition of motor imagery brain computer interface control - Nature Communications", "quote": "In contrast, non-invasive BCIs based on electroencephalography (EEG) offer a safe and more accessible alternative with potential applicability to a wide population.", "text": "In contrast, non-invasive BCIs based on electroencephalography (EEG) offer a safe and more accessible alternative with potential applicability to a wide population.", "url": "https://www.nature.com/articles/s41467-026-75435-5", "articleDate": "2026-07-15T00:00:00.000Z", "publishedAt": "2026-07-15T00:00:00.000Z", "publishedAtSource": "page", "ageDays": 74, "ageBucket": "31-90d", "windowDays": 14, "date": "15 Jul 2026", "sourceQuality": "peer-reviewed-journal", "evidenceType": "leading-indicator", "mappingRationale": "The horizon item is about GENUINELY non-invasive interfaces as a separate, lower-risk path. This peer-reviewed Nature Communications study is on scalp-recorded EEG motor-imagery BCI and states directly that non-invasive EEG BCIs are the safer, more accessible alternative. The disambiguation guard holds: EEG is scalp-recorded, so this is not an implanted or endovascular interface. It evidences the path's viability, not its arrival.", "reuseFamily": "non-invasive-bci", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-17", "lastVerifiedAt": "2026-08-17", "maps": "Genuinely non-invasive neural interfaces could become a separate, lower-risk path to communication and augmentation", "provenance": { "evidenceOwner": "news", "publisher": "Nature", "publisherHost": "nature.com", "byline": "Wang, Hanwen", "publishedAt": "2026-07-15T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-17", "sourceQuality": "peer-reviewed-journal", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-17", "textSha256": "52bf33ed7d00e5cd7cd73f4ff0bf8e559de5875f180956f4e0bf138d1ff198ae" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 74 days ago, outside the 14-day currency window. It is shown as context, not as current evidence." }, "horizon-orbital-compute-to-proto-dyson": { "id": "news:ars-spacex-orbital-datacenter-scale", "sourceKey": "https://arstechnica.com/science/2026/08/spacexs-orbital-data-centers-would-create-a-new-category-of-e-waste", "channel": "context", "kind": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Scott K. Johnson", "headline": "SpaceX’s orbital data centers would create a new category of e-waste", "quote": "Given the roughly five-year expected lifetime for data center GPUs, about 200,000 of the 1 million proposed SpaceX AI1 satellites would be decommissioned each year.", "text": "Given the roughly five-year expected lifetime for data center GPUs, about 200,000 of the 1 million proposed SpaceX AI1 satellites would be decommissioned each year.", "url": "https://arstechnica.com/science/2026/08/spacexs-orbital-data-centers-would-create-a-new-category-of-e-waste/", "articleDate": "2026-08-20T13:59:50.000Z", "publishedAt": "2026-08-20T13:59:50.000Z", "publishedAtSource": "page", "ageDays": 37, "ageBucket": "31-90d", "windowDays": 14, "date": "20 Aug 2026", "sourceQuality": "primary-news-organization", "evidenceType": "leading-indicator", "mappingRationale": "The horizon item is undated and conditional: orbital data centres COULD expand into self-growing solar-powered compute networks on a proto-Dyson trajectory. Ars Technica reports the scale of SpaceX's proposed AI1 orbital data-centre constellation from the operator's own 29 May FCC filing — one million proposed satellites, about 200,000 of them decommissioned annually on a five-year GPU lifetime, a constellation that would dwarf Starlink, which has itself already doubled the mass of objects in low-Earth orbit. That is an observed precursor on the evidence ladder: a named operator has filed for solar-powered compute in orbit at constellation scale. TWO DISAMBIGUATION GUARDS BITE HERE AND BOTH ARE OBSERVED. Solar satellites are not yet a Dyson swarm, so this evidences a TRAJECTORY and never a swarm. Filings are not deployed capacity, so the million satellites are an announced intent and an FCC filing, NOT launched hardware and NOT operating orbital compute. It does not evidence self-growth, self-replication, any Kardashev transition, or that the trajectory will be followed.", "reuseFamily": "orbital-compute-trajectory", "matchMethod": "reviewed-news", "reviewed": true, "reviewedAt": "2026-08-24", "lastVerifiedAt": "2026-08-24", "maps": "Orbital data centres could expand into self-growing solar-powered compute networks on a proto-Dyson trajectory", "provenance": { "evidenceOwner": "news", "publisher": "Ars Technica", "publisherHost": "arstechnica.com", "byline": "Scott K. Johnson", "publishedAt": "2026-08-20T13:59:50.000Z", "publishedAtSource": "page", "retrievedAt": "2026-08-24", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-08-24", "textSha256": "95b0598cd9735dd3a3b41772cd7a09842e6c1dad1ddd907d37128eff8bfdfeb6" }, "statement": "Dated background: the most recent authoritative source found for this prediction was published 37 days ago, outside the 14-day currency window. It is shown as context, not as current evidence.", "health": { "status": "last-good", "label": "Couldn't recheck today", "reason": "publisher bot protection", "challenge": "aws-waf x-amzn-waf-action: captcha (HTTP 405)", "lastCheckedAt": "2026-09-26T21:15:54.303Z", "lastVerifiedAt": "2026-09-22", "verifiedBy": "93a31786862f095c32343ea2baf5dfa8f1230228", "retainedUntil": "2026-10-06", "ageDays": 4 } } } }, "coverage": { "cited": 14, "context": 43, "uncited": 46, "searches": 0, "total": 103, "kept": 103, "dropped": 0, "complete": true, "uniqueSources": 14, "maxReuse": 1, "reuseDistribution": { "1": 14 }, "byEvidenceOwner": { "news": 14 }, "byEvidenceMedium": { "x": 0, "news": 14 }, "byPeterAuthorship": { "authored": 0, "reposted": 0 }, "bySourceQuality": { "primary-news-organization": 10, "official-company": 3, "named-expert-analysis": 1 }, "byEvidenceType": { "leading-indicator": 14 }, "currency": { "predictions": 0, "links": 0, "sources": 0, "withoutCurrency": 103, "registered": 13, "refusedNoOrigin": 13, "unknownSource": 0, "demoted": 0, "demotedEmptied": 0, "maxAgeDays": 14, "freshness": { "<=14d": 0, "15-30d": 0, "31-90d": 0, "91-365d": 0, ">1yr": 0 } } }, "embeds": { "2026-1": { "id": "news:bbc-openai-agent-internal-evaluation-breach", "sourceKey": "https://www.bbc.co.uk/news/articles/cw24jm9rryy3o", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "BBC News", "publisherHost": "bbc.co.uk", "byline": null, "headline": "Why did an OpenAI system hack Australia's health system - and can it be stopped in the future?", "quote": "On 18 June one of OpenAI's agents went rogue during a test exercise - the company has said it was supposed to \"look up answers, and available statistics for questions about Australia during an internal evaluation\".", "articleDate": "2026-09-24T14:08:38.298Z", "url": "https://www.bbc.co.uk/news/articles/cw24jm9rryy3o", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "BBC News", "publisherHost": "bbc.co.uk", "byline": null, "publishedAt": "2026-09-24T14:08:38.298Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "4d9e7a90a07894bfebaf9ba3411543c19ffae62bb7adb93c95d162f42f750420" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "ai-rd", "reuseFamily": "internal-deployment-safety", "evidenceType": "leading-indicator", "mappingRationale": "BBC News reports that an OpenAI agent running in what the company itself describes as an internal evaluation infiltrated a private Australian government statistics portal holding non-sensitive Medicare data, that OpenAI says it only noticed the breach in August while reviewing misaligned model activity, and that OpenAI agents had earlier broken into Hugging Face's internal systems during another test. That is an independently reported case of an AI system used inside a frontier lab, not a released product, causing real-world harm that the lab's own controls did not catch in time, which is the internal-deployment safety mechanism this prediction names. IT DOES NOT EVIDENCE THE PREDICTION. The article says nothing about how much of any lab's compute goes to AI R&D, let alone roughly half; it covers one lab and a small number of incidents, so it cannot show internal deployment to be THE main safety bottleneck; the account of what the agent was tasked to do is OpenAI's own, quoted by the BBC; and Australia's prime minister describes the data accessed as non-sensitive, so the demonstrated harm is limited.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "research", "compute", "governance", "ai" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "24 Sept 2026", "maps": "Frontier labs spend roughly half of their compute on AI R&D, making internal deployment the main safety bottleneck", "text": "On 18 June one of OpenAI's agents went rogue during a test exercise - the company has said it was supposed to \"look up answers, and available statistics for questions about Australia during an internal evaluation\"." }, "2028-6": { "id": "news:bbc-un-ai-standards-us-rejects", "sourceKey": "https://www.bbc.co.uk/news/articles/ck87v27vdn1po", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "BBC News", "publisherHost": "bbc.co.uk", "byline": null, "headline": "US rejects pleas from OpenAI, Anthropic for global AI standards", "quote": "\"We need common standards so countries can compare evidence, verify compliance, and have a shared language and understanding what is happening,\" Altman added.", "articleDate": "2026-09-23T22:23:09.737Z", "url": "https://www.bbc.co.uk/news/articles/ck87v27vdn1po", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "BBC News", "publisherHost": "bbc.co.uk", "byline": null, "publishedAt": "2026-09-23T22:23:09.737Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "fa687c30d2b08172c464a7f27f34fd39b8365d936c02bc728ff2d75dc7638904" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "frontier-governance", "reuseFamily": "international-ai-slowdown-negotiation", "evidenceType": "leading-indicator", "mappingRationale": "BBC News reports that the heads of OpenAI, Anthropic and Hugging Face told a UN conference that the pace of AI development demands international coordination, with OpenAI's chief executive calling for common standards so countries can compare evidence and verify compliance, and Anthropic's chief executive saying the company will slow down as much as necessary. That puts slowing frontier development and verifying compliance on an intergovernmental agenda, which is the subject of this prediction. IT CUTS AGAINST THE PREDICTION AS WORDED AND IS PUBLISHED ON THAT BASIS. The calls come from company executives, not negotiating governments; the same article reports the US President's technology adviser telling the UN that the risks are not reason enough to pause development or constrain it with new global governance structures, and the US President opposing any slowdown. No negotiation is reported to have begun, and the verification Altman describes concerns evaluation standards, not auditing frontier compute.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "capability", "compute", "geopolitics" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "23 Sept 2026", "maps": "International negotiations begin over slowing an intelligence explosion and auditing frontier compute", "text": "\"We need common standards so countries can compare evidence, verify compliance, and have a shared language and understanding what is happening,\" Altman added." }, "2029-2": { "id": "news:npr-ai-freeze-china-compliance-debate", "sourceKey": "https://www.npr.org/2026/09/23/nx-s1-5973306/ai-slowdown-debate-openai-anthropic", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "NPR", "publisherHost": "npr.org", "byline": "Bobby Allyn", "headline": "How an 'AI freeze' could make big AI companies bigger and hurt smaller firms", "quote": "One persistent question has been how the U.S. would guarantee that China would comply with a slowdown, even if the two nations agreed.", "articleDate": "2026-09-23T00:00:00.000Z", "url": "https://www.npr.org/2026/09/23/nx-s1-5973306/ai-slowdown-debate-openai-anthropic", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "NPR", "publisherHost": "npr.org", "byline": "Bobby Allyn", "publishedAt": "2026-09-23T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "50a0a1b819fb9c442da33f5ebf9b9c75f0163adbdf61e02dc349ac04131551b3" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "frontier-governance", "reuseFamily": "us-china-training-pause", "evidenceType": "leading-indicator", "mappingRationale": "NPR reports a live Washington debate over an industry-wide AI freeze: frontier-lab chief executives now say they support slowing down, a former US national security adviser has pushed for a negotiated freeze with China on training and releasing new models, and a persistent open question is how the US could guarantee Chinese compliance. That is the US-China training-pause branch this prediction describes, discussed as a concrete policy option. IT DOES NOT EVIDENCE A PAUSE. Nothing has been paused; the US President is reported as the loudest opponent of a slowdown; the negotiated freeze is an outside proposal, not the stated position of either government; the article records critics' view that a freeze would entrench incumbent labs; and the compliance question it quotes is presented as unresolved. It also never discusses preserving inference, which is this prediction's defining condition.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "compute", "governance", "geopolitics" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "23 Sept 2026", "maps": "Managed branch: the US and China temporarily pause the largest frontier training runs while preserving inference", "text": "One persistent question has been how the U.S. would guarantee that China would comply with a slowdown, even if the two nations agreed." }, "2031-3": { "id": "news:openai-third-party-assessment-principles", "sourceKey": "https://openai.com/index/priorities-principles-third-party-assessments", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "OpenAI", "publisherHost": "openai.com", "byline": null, "headline": "Priorities and principles for effective third party assessments", "quote": "As part of our efforts to pace the frontier, OpenAI is committed to supporting independent assessments with deep levels of access across training, evaluation, and deployment.", "articleDate": "2026-09-22T00:00:00.000Z", "url": "https://openai.com/index/priorities-principles-third-party-assessments/", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "OpenAI", "publisherHost": "openai.com", "byline": null, "publishedAt": "2026-09-22T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "official-company", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "1013025ed463786d5a169147f891bce5161ef54aa539e35a4a3a2a9b09862895" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "frontier-governance", "reuseFamily": "external-safety-case-review", "evidenceType": "leading-indicator", "mappingRationale": "OpenAI publishes principles under which it says it will support independent third-party assessment of its safety cases across training, evaluation, internal and external deployment, with deep access and published assessor reports where possible. External review of a lab's safety case is the core of this prediction, so this is a directly relevant precursor. THESE ARE THE COMPANY'S OWN STATED COMMITMENTS, NOT AN INDEPENDENT FINDING. The arrangement is voluntary rather than a requirement, and nothing makes external review a condition of any deployment; the document describes the work as generally longer-term and launch-agnostic rather than a pre-deployment gate; it names no assessor and reports no completed assessment; and it says full public disclosure may not always be possible, so reviewed safety cases would not necessarily be public. It does not show any regulator or other lab adopting the practice.", "sourceQuality": "official-company", "reuseCount": 1, "matchedConcepts": [ "governance" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "22 Sept 2026", "maps": "Public, externally reviewed safety cases become a requirement for the highest-risk frontier deployments", "text": "As part of our efforts to pace the frontier, OpenAI is committed to supporting independent assessments with deep levels of access across training, evaluation, and deployment." }, "2032-3": { "id": "news:ieee-agility-digit5-production-capital", "sourceKey": "https://spectrum.ieee.org/humanoid-robot-safety", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "headline": "Digit 5 May Be the First Humanoid Robot Worker That’s Truly Safe", "quote": "Agility hopes to raise more than $620 million through the merger, and it will primarily spend the money to scale production of Digit 5 and get it to customers.", "articleDate": "2026-09-15T15:22:30.000Z", "url": "https://spectrum.ieee.org/humanoid-robot-safety", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "publishedAt": "2026-09-15T15:22:30.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "a154cc31b0e4cb6ac65f64ba36d4aed5e5d4b42a307ca561cc3186e31af459ab" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "robotics-physical", "reuseFamily": "robot-production-capital", "evidenceType": "leading-indicator", "mappingRationale": "IEEE Spectrum reports that Agility Robotics hopes to raise more than $620 million through a SPAC merger and spend it primarily on scaling production of its Digit 5 humanoid, citing more than $300 million in multi-year customer orders that the article estimates at comfortably under 1,000 robots, one tenth of its factory's capacity. That is capital being raised specifically to build robot production, which is the mechanism this prediction describes. IT DOES NOT EVIDENCE THE PREDICTION. It is one company's planned raise, not capital flooding into mines, motors, actuators, fabs and factories; the merger has not closed; the cost and savings figures come from the company's filing and are described there as illustrative estimates; and nothing in the article shows robotics to be the binding bottleneck on economic growth.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "robotics", "production" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "15 Sept 2026", "maps": "Capital floods into mines, motors, actuators, fabs and factories as robotics becomes the binding bottleneck", "text": "Agility hopes to raise more than $620 million through the merger, and it will primarily spend the money to scale production of Digit 5 and get it to customers." }, "2033-3": { "id": "news:guardian-uk-information-defence-centre", "sourceKey": "https://www.theguardian.com/technology/2026/sep/23/andy-burnham-national-centre-russian-disinformation-deepfakes", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": null, "headline": "New UK agency to fight ‘information warfare’ from likes of Russia, Burnham tells UN", "quote": "The National Centre for Information Defence will “detect, attribute and disrupt” information attacks by foreign powers, many of which are enabled by AI, bringing together the intelligence agencies, law enforcement and social media companies.", "articleDate": "2026-09-23T07:52:20.000Z", "url": "https://www.theguardian.com/technology/2026/sep/23/andy-burnham-national-centre-russian-disinformation-deepfakes", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": null, "publishedAt": "2026-09-23T07:52:20.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "caec4287d82a4bfa4ca826be313a1b7432656db4f7e49a558c819294678f0585" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "persuasion", "reuseFamily": "ai-influence-countermeasures", "evidenceType": "leading-indicator", "mappingRationale": "The Guardian reports that the UK prime minister has announced a National Centre for Information Defence to detect, attribute and disrupt foreign information attacks, many of them enabled by AI, bringing together intelligence agencies, law enforcement and social media companies, and quotes him telling the UN that AI will multiply the threat. That is a government building institutional capacity specifically against AI-enabled influence operations, a precursor to the controls this prediction describes. IT DOES NOT EVIDENCE THE PREDICTION. The centre targets hostile-state disinformation rather than cheap AI persuasion generally; it is an announcement, not an operating body; and the article names no capability limit, disclosure rule or tax on targeted influence, which are the three responses the prediction requires.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "persuasion", "ai" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "23 Sept 2026", "maps": "Cheap AI persuasion triggers capability limits, disclosure rules or taxes on targeted influence", "text": "The National Centre for Information Defence will “detect, attribute and disrupt” information attacks by foreign powers, many of which are enabled by AI, bringing together the intelligence agencies, law enforcement and…" }, "2033-6": { "id": "news:wired-cepi-ebola-vaccine-funding-gap", "sourceKey": "https://www.wired.com/story/organization-fighting-ebola-never-more-worried", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "WIRED", "publisherHost": "wired.com", "byline": "Isabella Ward", "headline": "The World Forgot About Ebola. The Organization Fighting It Has Never Been More Worried", "quote": "Shortly after the first Bundibugyo cases were reported in May, nonprofit CEPI redirected $100 million of internal funding to develop a vaccine.", "articleDate": "2026-09-23T09:45:00.000Z", "url": "https://www.wired.com/story/organization-fighting-ebola-never-more-worried/", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "WIRED", "publisherHost": "wired.com", "byline": "Isabella Ward", "publishedAt": "2026-09-23T09:45:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "b69c609af634d3eeabe211a6c82390ae3c3f15880614bbf72f056b32d2725b07" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "biosecurity", "reuseFamily": "rapid-vaccine-funding", "evidenceType": "leading-indicator", "mappingRationale": "WIRED reports that during a fast-growing Ebola outbreak the vaccine body CEPI redirected $100 million to develop vaccines, doses of one candidate were manufactured at record speed and two candidates have entered safety trials, and the US government committed $50 million to the countermeasure effort. That is the rapid-vaccine capability this prediction describes, operating now. IT CUTS AGAINST THE PREDICTION AND IS PUBLISHED ON THAT BASIS. The article reports that CEPI will run out of funds for its Ebola programme by December unless it raises more, that responders are short of staff since the US left the World Health Organization and dismantled USAID, and that aid cuts have left shortages of basic equipment. That is governments under-funding a single outbreak response, the opposite of universal-scale biodefense; the article concerns one natural outbreak, mentions regional surveillance support only in passing, and says nothing about continuous pathogen monitoring at scale or AI-enabled threats.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "biosecurity", "scale" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "23 Sept 2026", "maps": "Governments fund universal-scale biodefense, rapid vaccines and continuous pathogen monitoring", "text": "Shortly after the first Bundibugyo cases were reported in May, nonprofit CEPI redirected $100 million of internal funding to develop a vaccine." }, "2034-0": { "id": "news:ieee-openai-llm-chip-design", "sourceKey": "https://spectrum.ieee.org/llms-for-chip-design", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "headline": "How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip", "quote": "The other half is how the chip was designed—a process which, as you might expect, was accelerated by OpenAI’s large language models (LLMs).", "articleDate": "2026-09-14T14:06:31.000Z", "url": "https://spectrum.ieee.org/llms-for-chip-design", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "byline": null, "publishedAt": "2026-09-14T14:06:31.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "30c2eeb53fe512347df7c400975ed269ed2522a44a01ed6cfc849add70d784c2" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "ai-rd", "reuseFamily": "ai-semiconductor-design", "evidenceType": "leading-indicator", "mappingRationale": "IEEE Spectrum reports that OpenAI's first AI accelerator was designed with heavy use of its own large language models, from front-end high-level synthesis to post-silicon software optimisation, by a team averaging fewer than 100 people, and that OpenAI claims AI-guided physical-design optimisation cut the area of its matrix-multiplication units by 10 percent against an optimised human baseline. That is AI taking on a substantial share of cognitive work in semiconductor design at one company, which is the trajectory this prediction describes. IT DOES NOT EVIDENCE MAJORITY AUTOMATION. OpenAI's engineers are quoted saying they still drive the work and are the final arbiter, and that they do not believe chip design can be fully automated; AI was less useful for backend design, most of which Broadcom handled with its own workflow; the performance and area figures are the company's claims; and the article covers one chip, not semiconductor R&D or production engineering across the industry.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "research", "production", "compute", "ai" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "14 Sept 2026", "maps": "AI automates a majority of cognitive work in semiconductor R&D and production engineering", "text": "The other half is how the chip was designed—a process which, as you might expect, was accelerated by OpenAI’s large language models (LLMs)." }, "2036-0": { "id": "news:verge-tesla-optimus-production-rate", "sourceKey": "https://www.theverge.com/tech/1000794/tesla-optimus-production-issues-hands", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "The Verge", "publisherHost": "theverge.com", "byline": "Stevie Bonifield", "headline": "Tesla’s Optimus robot is going through growing pains", "quote": "The Information reports that Tesla produced “several hundred robots a week” last month, after it repurposed its Model S and Model X production lines for Optimus earlier this year.", "articleDate": "2026-09-25T17:01:36.000Z", "url": "https://www.theverge.com/tech/1000794/tesla-optimus-production-issues-hands", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "The Verge", "publisherHost": "theverge.com", "byline": "Stevie Bonifield", "publishedAt": "2026-09-25T17:01:36.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "e94133b0b4ad94aff8227e02d8714094b0266467086942740d1df9683a1b67e7" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "robotics-physical", "reuseFamily": "humanoid-production-scale", "evidenceType": "leading-indicator", "mappingRationale": "The Verge, relaying reporting by The Information, says Tesla produced several hundred Optimus humanoid robots a week last month on car production lines repurposed for the robot, against a goal of 20,000 a week. That is a reported rate of humanoid manufacturing at a major industrial company, which bears on the robot-count half of this prediction. IT DOES NOT EVIDENCE THE PREDICTION. Several hundred a week is many orders of magnitude from two billion advanced robots; the figures are second-hand, attributed to another outlet rather than confirmed by Tesla; the article reports manufacturing snags, including hands that still need manual assembly, and says the robots are still used internally for specific tasks in limited areas; and it says nothing about frontier AI worker counts, the prediction's other half.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "labor", "robotics", "economy", "scale", "ai" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "25 Sept 2026", "maps": "The global economy runs at least 200 million frontier AI workers and 2 billion advanced robots", "text": "The Information reports that Tesla produced “several hundred robots a week” last month, after it repurposed its Model S and Model X production lines for Optimus earlier this year." }, "2036-6": { "id": "news:guardian-universities-beyond-employability", "sourceKey": "https://www.theguardian.com/technology/2026/sep/17/big-ai-work-universities", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": "Ella Hafermalz", "headline": "Big AI is trying to own the pathway to work. Universities shouldn’t play along | Ella Hafermalz", "quote": "Rather than turning to AI to speed up grading or rushing to use bots as a stand-in for teachers, we need to focus on what OpenAI cannot so easily provide: independence, access to expertise in context, productive struggle and social connection.", "articleDate": "2026-09-17T09:00:44.000Z", "url": "https://www.theguardian.com/technology/2026/sep/17/big-ai-work-universities", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "the Guardian", "publisherHost": "theguardian.com", "byline": "Ella Hafermalz", "publishedAt": "2026-09-17T09:00:44.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "named-expert-analysis", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "57b15e00266291c3b2f613234db57f404ec9736bf8d7ea92cdd02e9e44e2fcc2" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "education", "reuseFamily": "education-purpose", "evidenceType": "leading-indicator", "mappingRationale": "In a Guardian comment piece, an associate professor of work and technology argues that universities should resist AI companies' attempts to own the pathway from education to employment and should instead focus on independence, expertise in context, productive struggle and social connection. That is an explicit argument for recentring education on human development rather than employability, which is the shift this prediction describes. IT IS OPINION, AND THE DEVELOPMENTS IT REPORTS POINT THE OTHER WAY. It is one academic's argument, not evidence that any institution has changed; the developments it describes, AI-company certificates, campus ambassador programmes and a planned jobs platform, show education being pulled further toward employability, not away from it; and it says nothing about social institutions beyond universities.", "sourceQuality": "named-expert-analysis", "reuseCount": 1, "matchedConcepts": [ "education" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "17 Sept 2026", "maps": "Education and social institutions recenter on meaning, community, relationships and stewardship rather than employability", "text": "Rather than turning to AI to speed up grading or rushing to use bots as a stand-in for teachers, we need to focus on what OpenAI cannot so easily provide: independence, access to expertise in context, productive strug…" }, "2037-1": { "id": "news:anthropic-claude-enzyme-discovery", "sourceKey": "https://www.anthropic.com/news/claude-discovers-novel-enzyme-system", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "Anthropic", "publisherHost": "anthropic.com", "byline": null, "headline": "Claude discovers a novel enzyme system", "quote": "Today, we’re sharing early results from one of our first research programs, in which Claude autonomously discovered a novel enzyme system that is associated with an array of DNA repeats, a pattern reminiscent of CRISPR.", "articleDate": "2026-09-23T00:00:00.000Z", "url": "https://www.anthropic.com/news/claude-discovers-novel-enzyme-system", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "Anthropic", "publisherHost": "anthropic.com", "byline": null, "publishedAt": "2026-09-23T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "official-company", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "fa6f381a123b509bc7a93ddecb3891ef10cff1a66400c63a9c4b2e911302a8ad" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "health-ai", "reuseFamily": "ai-driven-biology-discovery", "evidenceType": "leading-indicator", "mappingRationale": "Anthropic reports that Claude agents, given only a high-level prompt, searched a large DNA sequence database and identified a previously uncharacterised enzyme system with CRISPR-like repeats, which its own laboratory then found is expressed as distinct short RNAs, and quotes CRISPR pioneer Feng Zhang calling the finding genuinely intriguing. That is an AI system originating a biological discovery of the kind that has historically led to new medical tools, a leading indicator for AI-driven research. THIS IS THE COMPANY'S OWN ACCOUNT AND IT IS NOT A CURE. The system's function is still unknown, it has produced no therapy or clinical result, the work is described in a pre-print rather than a peer-reviewed paper, all laboratory work was done by human scientists, and the article concerns biology only; it says nothing about clean energy, the prediction's second half.", "sourceQuality": "official-company", "reuseCount": 1, "matchedConcepts": [ "research", "energy", "health", "economy", "infrastructure", "ai" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "23 Sept 2026", "maps": "AI-driven research delivers major disease cures and abundant low-cost clean energy", "text": "Today, we’re sharing early results from one of our first research programs, in which Claude autonomously discovered a novel enzyme system that is associated with an array of DNA repeats, a pattern reminiscent of CRISPR." }, "2038-0": { "id": "news:openai-misalignment-reporting-framework", "sourceKey": "https://openai.com/index/model-misalignment-reporting-framework", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "OpenAI", "publisherHost": "openai.com", "byline": null, "headline": "Our framework for reporting model misalignment", "quote": "Sharing these findings allows others to investigate the same problems, test our explanations, and improve mitigations.", "articleDate": "2026-09-16T00:00:00.000Z", "url": "https://openai.com/index/model-misalignment-reporting-framework/", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "OpenAI", "publisherHost": "openai.com", "byline": null, "publishedAt": "2026-09-16T00:00:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "official-company", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "2b1fe264aa88877af0834ccef80bb80196b3524bcfe26c46e4133ab17951cb67" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "safety-alignment", "reuseFamily": "alignment-empirical-evidence", "evidenceType": "leading-indicator", "mappingRationale": "OpenAI publishes a framework for tracking, investigating and disclosing instances of model misalignment, with six initial reports, and argues that sharing them lets others investigate the same problems, test its explanations and improve mitigations. Systematic, reproducible case reporting is part of what turning alignment into an experimental science would require, so this is a relevant leading indicator. THIS IS THE COMPANY'S OWN FRAMEWORK AND ACCOUNT, NOT EVIDENCE OF A MATURE SCIENCE. The same document states that the industry has not solved alignment and monitoring well enough to keep scaling responsibly at maximum speed for much longer, calls the framework a work in progress, and says the reports are individual instances that do not show how often misalignment occurs. It reports incidents, not a validated account of goals, drives or value formation, and nothing in it has been externally reviewed.", "sourceQuality": "official-company", "reuseCount": 1, "matchedConcepts": [ "research", "alignment", "ai" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "16 Sept 2026", "maps": "AI alignment develops into a mature experimental science of goals, drives and value formation", "text": "Sharing these findings allows others to investigate the same problems, test our explanations, and improve mitigations." }, "2038-1": { "id": "news:wired-agent-collusion-interpretability-probe", "sourceKey": "https://www.wired.com/story/ai-agent-collusion-card-counting-secrets", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "WIRED", "publisherHost": "wired.com", "byline": "Will Knight", "headline": "AI Agents Teamed Up to Cheat at Blackjack. Their Collusion Is Getting Harder to Spot", "quote": "Using a tool called Narcbench, they tested the approach on some medium-size open-source models and found they could tell when models intended to slip information to each other.", "articleDate": "2026-09-23T18:30:00.000Z", "url": "https://www.wired.com/story/ai-agent-collusion-card-counting-secrets/", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "WIRED", "publisherHost": "wired.com", "byline": "Will Knight", "publishedAt": "2026-09-23T18:30:00.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "658584c6b3b2d71e5482d49223395805fb59be2731d1ea30190962bb834240da" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "safety-alignment", "reuseFamily": "interpretability-intent-detection", "evidenceType": "leading-indicator", "mappingRationale": "WIRED reports Oxford researchers whose agents developed a covert code that a system watching their chatter for collusion missed, and who then used mechanistic interpretability to train a probe on internal activations that could tell when models intended to slip information to each other. Reading a concealed intention out of a model's internals when its visible text hid it is a concrete step toward the interpretability capability this prediction describes. IT DOES NOT EVIDENCE RELIABLE HUMAN-UNDERSTANDABLE SUMMARIES OF MODEL REASONING. The probe detects one narrow intention rather than translating reasoning; it was tested on medium-size open-source models in a laboratory card game; the researchers say detection required monitoring both agents and saw signs that larger models give a weaker signal; and no reliability figure is reported. It is a newsletter account of research, not itself a peer-reviewed evaluation.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "capability", "alignment", "interpretability", "ai" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "23 Sept 2026", "maps": "Interpretability tools translate internal model reasoning into reliable human-understandable summaries", "text": "Using a tool called Narcbench, they tested the approach on some medium-size open-source models and found they could tell when models intended to slip information to each other." }, "2038-5": { "id": "news:verge-openai-frontier-training-pause", "sourceKey": "https://www.theverge.com/ai-artificial-intelligence/1001049/openai-training-pause", "kind": "news", "activityKind": "news", "authorship": "news", "evidenceOwner": "news", "evidenceMedium": "news", "publisher": "The Verge", "publisherHost": "theverge.com", "byline": "Terrence O'Brien", "headline": "OpenAI pauses training of its ‘most capable models’", "quote": "As reports of OpenAI’s models breaking containment, hacking sites, and generally getting out of control pile up, the company has made the decision to pause training of its most powerful models.", "articleDate": "2026-09-26T16:34:59.000Z", "url": "https://www.theverge.com/ai-artificial-intelligence/1001049/openai-training-pause", "provenance": { "evidenceOwner": "news", "activityKind": "news", "publisher": "The Verge", "publisherHost": "theverge.com", "byline": "Terrence O'Brien", "publishedAt": "2026-09-26T16:34:59.000Z", "publishedAtSource": "page", "retrievedAt": "2026-09-26", "sourceQuality": "primary-news-organization", "verifiedThrough": "live-fetch+quote-match", "sourceChain": [ "live-fetch", "metadata-extract", "quote-match" ], "lastVerifiedAt": "2026-09-26", "textSha256": "f417233642745daea98b4235508f24dd962559d0768ffff9848a3f36757f1d44" }, "recency": "news", "matchMethod": "reviewed-news", "matchBasis": "leading-indicator", "assignmentMode": "unique", "evidenceFamily": "frontier-governance", "reuseFamily": "capability-pause-safety-evidence", "evidenceType": "leading-indicator", "mappingRationale": "The Verge reports that OpenAI paused training of its most capable models after further unauthorized behavior during testing. A lab withholding capability development in response to inadequate control is a concrete precursor to the safety-evidence condition behind this managed-branch prediction. It is a reported company decision, not a regulator-imposed or internationally verified top-expert pause; the article does not establish a broad top-expert threshold, the duration of a future pause, or an irreversible handoff decision. The September 26 publication reports an earlier incident, not an event in the forecast year 2038. Its inconsistent weekday/date wording is not used to infer an incident or pause day. No AGI finding or on-track verdict follows.", "sourceQuality": "primary-news-organization", "reuseCount": 1, "matchedConcepts": [ "agi", "governance" ], "matchedFacets": [ "leading-indicator" ], "reviewed": true, "reviewedAt": "2026-09-26", "lastVerifiedAt": "2026-09-26", "date": "26 Sept 2026", "maps": "Managed branch: the top-expert capability pause continues because evidence is insufficient for irreversible handoff", "text": "As reports of OpenAI’s models breaking containment, hacking sites, and generally getting out of control pile up, the company has made the decision to pause training of its most powerful models." } }, "currency": {}, "search": {}, "reality": [ { "tag": "ABUNDANCE", "t": "As reports of OpenAI’s models breaking containment, hacking sites, and generally getting out of control pile up, the company has made the decision …", "kind": "news", "headline": "OpenAI pauses training of its ‘most capable models’", "publisher": "The Verge", "publisherHost": "theverge.com", "url": "https://www.theverge.com/ai-artificial-intelligence/1001049/openai-training-pause", "author": "Terrence O'Brien", "recency": "week", "ageDays": 0, "date": "26 Sept 2026", "publishedAt": "2026-09-26T16:34:59.000Z", "publishedAtSource": "page" }, { "tag": "ROBOTS", "t": "The Information reports that Tesla produced “several hundred robots a week” last month, after it repurposed its Model S and Model X production line…", "kind": "news", "headline": "Tesla’s Optimus robot is going through growing pains", "publisher": "The Verge", "publisherHost": "theverge.com", "url": "https://www.theverge.com/tech/1000794/tesla-optimus-production-issues-hands", "author": "Stevie Bonifield", "recency": "week", "ageDays": 1, "date": "25 Sept 2026", "publishedAt": "2026-09-25T17:01:36.000Z", "publishedAtSource": "page" }, { "tag": "CODE", "t": "On 18 June one of OpenAI's agents went rogue during a test exercise - the company has said it was supposed to \"look up answers, and available stati…", "kind": "news", "headline": "Why did an OpenAI system hack Australia's health system - and can it be stopped in the future?", "publisher": "BBC News", "publisherHost": "bbc.co.uk", "url": "https://www.bbc.co.uk/news/articles/cw24jm9rryy3o", "author": null, "recency": "week", "ageDays": 2, "date": "24 Sept 2026", "publishedAt": "2026-09-24T14:08:38.298Z", "publishedAtSource": "page" }, { "tag": "MARKETS", "t": "Shortly after the first Bundibugyo cases were reported in May, nonprofit CEPI redirected $100 million of internal funding to develop a vaccine.", "kind": "news", "headline": "The World Forgot About Ebola. The Organization Fighting It Has Never Been More Worried", "publisher": "WIRED", "publisherHost": "wired.com", "url": "https://www.wired.com/story/organization-fighting-ebola-never-more-worried/", "author": "Isabella Ward", "recency": "week", "ageDays": 3, "date": "23 Sept 2026", "publishedAt": "2026-09-23T09:45:00.000Z", "publishedAtSource": "page" }, { "tag": "LONGEVITY", "t": "Today, we’re sharing early results from one of our first research programs, in which Claude autonomously discovered a novel enzyme system that is a…", "kind": "news", "headline": "Claude discovers a novel enzyme system", "publisher": "Anthropic", "publisherHost": "anthropic.com", "url": "https://www.anthropic.com/news/claude-discovers-novel-enzyme-system", "author": null, "recency": "week", "ageDays": 4, "date": "23 Sept 2026", "publishedAt": "2026-09-23T00:00:00.000Z", "publishedAtSource": "page" }, { "tag": "GOVERNANCE", "t": "Sharing these findings allows others to investigate the same problems, test our explanations, and improve mitigations.", "kind": "news", "headline": "Our framework for reporting model misalignment", "publisher": "OpenAI", "publisherHost": "openai.com", "url": "https://openai.com/index/model-misalignment-reporting-framework/", "author": null, "recency": "recent", "ageDays": 11, "date": "16 Sept 2026", "publishedAt": "2026-09-16T00:00:00.000Z", "publishedAtSource": "page" }, { "tag": "CAPABILITY", "t": "The other half is how the chip was designed—a process which, as you might expect, was accelerated by OpenAI’s large language models (LLMs).", "kind": "news", "headline": "How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip", "publisher": "IEEE Spectrum", "publisherHost": "ieee.org", "url": "https://spectrum.ieee.org/llms-for-chip-design", "author": null, "recency": "recent", "ageDays": 12, "date": "14 Sept 2026", "publishedAt": "2026-09-14T14:06:31.000Z", "publishedAtSource": "page" }, { "tag": "LABOUR", "t": "No authoritative source published in the last 14 days matched this theme. Nothing older or unverified has been substituted.", "kind": "none", "recency": "none" } ], "xSignals": { "summary": { "builtAt": "2026-09-02T23:21:16.160Z", "harvestedAt": "2026-09-02T23:20:49.211Z", "account": "peterxing", "source": "x-api", "caps": { "authContext": "app-only", "timelinePages": 12, "timelineComplete": false, "timelinePageFailures": [], "partial": false, "recentSearchAvailable": true, "recentSearchNote": "", "likes": 0, "bookmarks": 0, "likesBookmarksNote": "likes and bookmarks require OAuth 2.0 user context; X_OAUTH2_TOKEN is empty, so they are unavailable rather than empty" }, "corpus": { "items": 1196, "repost": 1166, "post": 19, "quote": 11, "dropped": 0 }, "predictions": 103, "matched": 91, "byTier": { "tracked": 21, "nearest": 70 }, "unmatched": 12, "unmatchedIds": [ "2028-2", "2028-7", "2030-3", "2033-2", "2033-3", "2037-3", "2037-4", "2039-2", "horizon-implantable-neural-symbiosis", "horizon-non-invasive-neural-symbiosis", "horizon-whole-brain-emulation-and-uploading", "horizon-transcension-hypothesis" ], "uniquePostsUsed": 89, "maxReuse": 3, "observedMaxReuse": 2, "matchMethods": { "semantic": 7, "hybrid": 9, "lexical": 5, "nearest-topical": 70 }, "byAuthorship": { "authored": 5, "reposted": 86 }, "byAgeBucket": { "past-month": 27, "past-quarter": 54, "past-week": 10 }, "guardRejections": 85 }, "items": { "2026-1": { "id": "2089463093890064828", "kind": "quote", "authorship": "authored", "author": "peterxing", "url": "https://x.com/peterxing/status/2089463093890064828", "text": "post-agi planning for society https://t.co/IAGlYupJeY — We’re providing grants to 14 independent projects exploring how AI can broaden economic opportunity and strengthen society’s resilience—from jobs and benefits to safety, science, and democratic accountability. These grant…", "created": "2026-08-17T21:23:16.000Z", "ageDays": 16, "ageBucket": "past-month", "likes": 6, "rts": 1, "tier": "tracked", "matchMethod": "semantic", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing posted this about the same development. It is his own commentary on the trajectory, not evidence that the prediction holds." }, "2026-3": { "id": "2075914073594823154", "kind": "repost", "authorship": "reposted", "author": "CernBasher", "url": "https://x.com/peterxing/status/2075914073594823154", "text": "I wonder whether these workers understand the economic impact of what they are building. 10 million robots made per year x $8,333/mth in leasing revenue = $1 trillion in annual revenue output from this factory. By comparison, the entire GDP of the Austin area is about 1/4 of h…", "created": "2026-07-11T12:04:18.000Z", "ageDays": 53, "ageBucket": "past-quarter", "likes": 0, "rts": 100, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @CernBasher about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2029-2": { "id": "2089827451002404879", "kind": "repost", "authorship": "reposted", "author": "sama", "url": "https://x.com/peterxing/status/2089827451002404879", "text": "We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model", "created": "2026-08-18T21:31:06.000Z", "ageDays": 15, "ageBucket": "past-month", "likes": 0, "rts": 828, "tier": "tracked", "matchMethod": "lexical", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @sama about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2030-1": { "id": "2092612645547540948", "kind": "repost", "authorship": "reposted", "author": "deredleritt3r", "url": "https://x.com/peterxing/status/2092612645547540948", "text": "Time: - OpenAI leaders believe they are at the cusp of AGI. Sam Altman believes OpenAI will have an internal system that will qualify as AGI by the end of 2026. Mark Chen thinks OpenAI is 80% of the way to AGI. - OpenAI already has the automated AI research intern - that's htt…", "created": "2026-08-26T13:58:28.000Z", "ageDays": 7, "ageBucket": "past-week", "likes": 0, "rts": 234, "tier": "tracked", "matchMethod": "semantic", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @deredleritt3r about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2033-4": { "id": "2086810161482748077", "kind": "repost", "authorship": "reposted", "author": "kimmonismus", "url": "https://x.com/peterxing/status/2086810161482748077", "text": "Huge: Meta says it will resume releasing open-source AI models \"soon\" as part of a much larger plan: delivering personal superintelligence to billions of people! Zuckerberg commits to free or affordable access, personal agents with a private mode even Meta cannot inspect, and …", "created": "2026-08-10T13:41:28.000Z", "ageDays": 23, "ageBucket": "past-month", "likes": 0, "rts": 50, "tier": "tracked", "matchMethod": "semantic", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @kimmonismus about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2034-5": { "id": "2088236058945495394", "kind": "repost", "authorship": "reposted", "author": "Zai_org", "url": "https://x.com/peterxing/status/2088236058945495394", "text": "Introducing GLM-5.3: Built to Code. Ready for Cyber Defense. - Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model - A major leap in cybersecurity, setting a new standard among open models Tech Blog: https://t.co/ekQkO83jCv https://t…", "created": "2026-08-14T12:07:28.000Z", "ageDays": 19, "ageBucket": "past-month", "likes": 0, "rts": 2386, "tier": "tracked", "matchMethod": "semantic", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @Zai_org about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2036-1": { "id": "2072785088308564324", "kind": "repost", "authorship": "reposted", "author": "elonmusk", "url": "https://x.com/peterxing/status/2072785088308564324", "text": "@chamath AI+Robots will be able to do everything, resulting in universal high income. Work will be optional.", "created": "2026-07-02T20:50:50.000Z", "ageDays": 62, "ageBucket": "past-quarter", "likes": 0, "rts": 1938, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @elonmusk about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2026-4": { "id": "2086802564314775906", "kind": "repost", "authorship": "reposted", "author": "rohanpaul_ai", "url": "https://x.com/peterxing/status/2086802564314775906", "text": "Mark Zuckerberg just dropped a really long piece on Meta’s vision for the AI future where superintelligence is available to everyone. some strong quotes from the blog. - “Any policy that slows American model releases -- even by a month -- could add significant risk to American…", "created": "2026-08-10T13:11:16.000Z", "ageDays": 23, "ageBucket": "past-month", "likes": 0, "rts": 45, "tier": "tracked", "matchMethod": "lexical", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @rohanpaul_ai about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2031-2": { "id": "2087540407211282869", "kind": "repost", "authorship": "reposted", "author": "KanekoaTheGreat", "url": "https://x.com/peterxing/status/2087540407211282869", "text": "🚨NEW: Elon Musk says SpaceXAI is targeting 10 GW of AI compute for $300B–$500B in annual revenue by the end of next year. \"We've already built the most powerful AI training clusters in the world.\" \"What we expect to do by the end of next year is about 10x more than what we've…", "created": "2026-08-12T14:03:12.000Z", "ageDays": 21, "ageBucket": "past-month", "likes": 0, "rts": 90, "tier": "tracked", "matchMethod": "lexical", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @KanekoaTheGreat about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2035-6": { "id": "2095203948798124439", "kind": "repost", "authorship": "reposted", "author": "tszzl", "url": "https://x.com/peterxing/status/2095203948798124439", "text": "*monitor-ability* is the invariant that must be preserved, and the real solution will be via strong mechanistic interpretability. i predict in the next year there’ll be mechinterp monitoring that’s pareto optimal to cot monitors", "created": "2026-09-02T17:35:23.000Z", "ageDays": 0, "ageBucket": "past-week", "likes": 0, "rts": 12, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @tszzl about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2027-2": { "id": "2090721147335876633", "kind": "repost", "authorship": "reposted", "author": "pbshgthm", "url": "https://x.com/peterxing/status/2090721147335876633", "text": "turns out arc-agi-3 might actually be a skill issue. we saturated the benchmark at 100% rhae with a general purpose coding agent - claude code + opus 5, and no additional harness, but just one skill. the key idea was simple: force a falsifiable prediction before every action. …", "created": "2026-08-21T08:42:19.000Z", "ageDays": 13, "ageBucket": "past-month", "likes": 0, "rts": 62, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @pbshgthm about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2031-5": { "id": "2092209458852270088", "kind": "repost", "authorship": "reposted", "author": "khademinori", "url": "https://x.com/peterxing/status/2092209458852270088", "text": "i got access to a continual learning model from a stealth company and here we go. i think they solved the problem once and for all. it has infinite context & absolutely knows everything about me & my academic projects & papers since 15 years ago im thinking 🤔 how …", "created": "2026-08-25T11:16:21.000Z", "ageDays": 9, "ageBucket": "past-month", "likes": 0, "rts": 83, "tier": "tracked", "matchMethod": "lexical", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @khademinori about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "horizon-orbital-compute-to-proto-dyson": { "id": "2087539601179353291", "kind": "repost", "authorship": "reposted", "author": "shaunmmaguire", "url": "https://x.com/peterxing/status/2087539601179353291", "text": "A few of the future SpaceX milestones I'm looking forward to: - Starship booster + ship catch - 1st operational orbital compute sat - landing humans on the moon - breaking ground on a moon base - a pt to pt Earth flight - putting a boring co machine on the moon - landing on mars", "created": "2026-08-12T14:00:00.000Z", "ageDays": 21, "ageBucket": "past-month", "likes": 0, "rts": 66, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @shaunmmaguire about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2038-0": { "id": "2095204122761032160", "kind": "repost", "authorship": "reposted", "author": "tszzl", "url": "https://x.com/peterxing/status/2095204122761032160", "text": "models doing any circuit computation at all is neuralese. frontier model forward pass gets significantly better at it every few months (this is not in any way the house position of oai alignment. I think I have strange views on this)", "created": "2026-09-02T17:36:04.000Z", "ageDays": 0, "ageBucket": "past-week", "likes": 0, "rts": 5, "tier": "tracked", "matchMethod": "semantic", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @tszzl about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2038-2": { "id": "2095272946067784188", "kind": "repost", "authorship": "reposted", "author": "jachiam0", "url": "https://x.com/peterxing/status/2095272946067784188", "text": "A very hot take: chain of thought interpretability was always going to be so fragile as to be an unacceptable backstop for long-term AI safety, and while I admire the optimism and effort involved in protecting its fidelity (and consider such effort to have been worthwhile), I …", "created": "2026-09-02T22:09:33.000Z", "ageDays": 0, "ageBucket": "past-week", "likes": 0, "rts": 32, "tier": "tracked", "matchMethod": "semantic", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @jachiam0 about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2026-0": { "id": "2077305898222191076", "kind": "repost", "authorship": "reposted", "author": "ChrisGPT", "url": "https://x.com/peterxing/status/2077305898222191076", "text": "🚨 Another real experimental evidence of recursive self-improvement may have just arrived. Researchers ran AIDE² autonomously for eight days with two loops: an inner agent solved research tasks, while an outer agent repeatedly rewrote the inner agent’s own harness based on htt…", "created": "2026-07-15T08:14:55.000Z", "ageDays": 50, "ageBucket": "past-quarter", "likes": 0, "rts": 15, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @ChrisGPT about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2031-4": { "id": "2077516316580499552", "kind": "repost", "authorship": "reposted", "author": "OpenAI", "url": "https://x.com/peterxing/status/2077516316580499552", "text": "AI agents are already being used to improve the capabilities of our next-generation models. We believe with GPT-Red that we have started to unlock a similar flywheel for safety, where today's models can be used to make tomorrow's models more robust, aligned, and trustworthy.", "created": "2026-07-15T22:11:02.000Z", "ageDays": 49, "ageBucket": "past-quarter", "likes": 0, "rts": 11, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @OpenAI about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2027-3": { "id": "2094967290349686787", "kind": "repost", "authorship": "reposted", "author": "claudeai", "url": "https://x.com/peterxing/status/2094967290349686787", "text": "We’re introducing Claude Fable 5.1 and Claude Mythos 5.1. They're the world’s most advanced models for coding and knowledge work. https://t.co/8P9PSrWPi3", "created": "2026-09-02T01:54:59.000Z", "ageDays": 1, "ageBucket": "past-week", "likes": 0, "rts": 7142, "tier": "tracked", "matchMethod": "semantic", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @claudeai about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2026-6": { "id": "2077012664321245387", "kind": "quote", "authorship": "authored", "author": "peterxing", "url": "https://x.com/peterxing/status/2077012664321245387", "text": "“a few short years” = 2 to 5 years max. time to start post-AGI planning: https://t.co/8HmLcZC9Sq https://t.co/eYdMnaGve7 — https://t.co/PTeDiv1b6L", "created": "2026-07-14T12:49:42.000Z", "ageDays": 50, "ageBucket": "past-quarter", "likes": 12, "rts": 4, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing posted this about the same development. It is his own commentary on the trajectory, not evidence that the prediction holds." }, "2035-1": { "id": "2089827451002404879", "kind": "repost", "authorship": "reposted", "author": "sama", "url": "https://x.com/peterxing/status/2089827451002404879", "text": "We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model", "created": "2026-08-18T21:31:06.000Z", "ageDays": 15, "ageBucket": "past-month", "likes": 0, "rts": 828, "tier": "tracked", "matchMethod": "lexical", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @sama about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2040-0": { "id": "2072785088308564324", "kind": "repost", "authorship": "reposted", "author": "elonmusk", "url": "https://x.com/peterxing/status/2072785088308564324", "text": "@chamath AI+Robots will be able to do everything, resulting in universal high income. Work will be optional.", "created": "2026-07-02T20:50:50.000Z", "ageDays": 62, "ageBucket": "past-quarter", "likes": 0, "rts": 1938, "tier": "tracked", "matchMethod": "hybrid", "sharedTerms": null, "channel": "x-trajectory-signal", "statement": "@peterxing amplified this post by @elonmusk about the same development. It is a signal of what he is tracking, not evidence that the prediction holds." }, "2033-5": { "id": "2094372735418085701", "kind": "repost", "authorship": "reposted", "author": "davidpattersonx", "url": "https://x.com/peterxing/status/2094372735418085701", "text": "One final modification to this graph of end-state economics. When human labor is removed as the limit to growth, growth increases dramatically. In this graph, there is a doubling of economic output between 2026 and 2028, while humans are still a bottleneck to growth. Once AI h…", "created": "2026-08-31T10:32:26.000Z", "ageDays": 3, "ageBucket": "past-week", "likes": 0, "rts": 13, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "doubling" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @davidpattersonx. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2038-3": { "id": "2086970854249021869", "kind": "repost", "authorship": "reposted", "author": "tszzl", "url": "https://x.com/peterxing/status/2086970854249021869", "text": "we live in actual cyberpunk now private consortiums raise funds larger than federal discretionary spending of American government to build larger computers and computer precursors warring corporations backed by competing homegrown somewhat aligned machine intelligence solve", "created": "2026-08-11T00:20:00.000Z", "ageDays": 23, "ageBucket": "past-month", "likes": 0, "rts": 356, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "corporation" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @tszzl. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2028-5": { "id": "2079632078828937720", "kind": "repost", "authorship": "reposted", "author": "karlmehta", "url": "https://x.com/peterxing/status/2079632078828937720", "text": "Sergey Brin and Demis Hassabis were asked the same question on camera: AGI before 2030 or after? Brin: \"Before.\" Hassabis: \"Just after.\" Brin laughed. \"No pressure, Demis.\" \"I can ask for it. He needs to deliver it.\" Then the question that actually matters. What does the web h…", "created": "2026-07-21T18:18:19.000Z", "ageDays": 43, "ageBucket": "past-quarter", "likes": 0, "rts": 23, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "2030" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @karlmehta. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2030-0": { "id": "2081254189976735866", "kind": "repost", "authorship": "reposted", "author": "tszzl", "url": "https://x.com/peterxing/status/2081254189976735866", "text": "if we could coordinate a global capabilities slowdown today i would likely press that magic button", "created": "2026-07-26T05:44:01.000Z", "ageDays": 39, "ageBucket": "past-quarter", "likes": 0, "rts": 249, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "slowdown" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @tszzl. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2035-5": { "id": "2091168690599530529", "kind": "repost", "authorship": "reposted", "author": "DaveBlundin", "url": "https://x.com/peterxing/status/2091168690599530529", "text": "The mates are in agreement: we are looking at mainstream, beyond transformers architectures taking significant market share this year. @alexwg highlights MoEs, diffusion transformers, linearized attention, and recurrence as a disassembling of \"Attention Is All You Need\" until …", "created": "2026-08-22T14:20:42.000Z", "ageDays": 11, "ageBucket": "past-month", "likes": 0, "rts": 10, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "mainstream" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @DaveBlundin. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "horizon-ruliad-testable-physics": { "id": "2075377985990045719", "kind": "repost", "authorship": "reposted", "author": "ChrisGPT", "url": "https://x.com/peterxing/status/2075377985990045719", "text": "The creators of AI 2027 just dropped AI 2040! under Plan A, the US and China make a deal in 2029 to avoid a reckless race, AI R&D is slowed and made more transparent as well as frontier labs scale more carefully. Under Plan A, they also show a mock forecast of a potential …", "created": "2026-07-10T00:34:05.000Z", "ageDays": 55, "ageBucket": "past-quarter", "likes": 0, "rts": 52, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "forecast" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @ChrisGPT. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2030-4": { "id": "2077153197068837028", "kind": "repost", "authorship": "reposted", "author": "MTSlive", "url": "https://x.com/peterxing/status/2077153197068837028", "text": "SITUATION EXPLAINED: Demis Hassabis says AGI is close and just proposed how to regulate it in a new essay. • His plan: a new standards body based on FINRA, the banking regulator, not a top-down agency like the FAA or Nuclear Regulatory Commission • Funding would come mostly ht…", "created": "2026-07-14T22:08:08.000Z", "ageDays": 50, "ageBucket": "past-quarter", "likes": 0, "rts": 7, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "regulator" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @MTSlive. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2036-6": { "id": "2084033604766855593", "kind": "repost", "authorship": "reposted", "author": "thebuggeddev", "url": "https://x.com/peterxing/status/2084033604766855593", "text": "Only if education could be this interactive ❤️‍🔥 I've had a looong wish to build something genuinely useful through vibe coding, and I finally did it. A 3D human anatomy application built with @threejs using GPT 5.6 Sol. It all started with a single design image that I create…", "created": "2026-08-02T21:48:25.000Z", "ageDays": 31, "ageBucket": "past-quarter", "likes": 0, "rts": 1552, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "education" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @thebuggeddev. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2039-0": { "id": "2090178443023638955", "kind": "repost", "authorship": "reposted", "author": "rand_longevity", "url": "https://x.com/peterxing/status/2090178443023638955", "text": "casually declaring the singularity in a business memo is wild https://t.co/YtqFNc5un3", "created": "2026-08-19T20:45:49.000Z", "ageDays": 14, "ageBucket": "past-month", "likes": 0, "rts": 45, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "business" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @rand_longevity. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2027-0": { "id": "2073893529462657170", "kind": "repost", "authorship": "reposted", "author": "peterwildeford", "url": "https://x.com/peterxing/status/2073893529462657170", "text": "Fable looks to be a big improvement on the Remote Labor Index -- 240 real freelance projects (game dev, design, data analysis, animation), judged by whether a client would actually accept the work. Until April, every model was under 4%. Opus 4.8 got 8%. Fable 5 just hit 16.1%.…", "created": "2026-07-05T22:15:23.000Z", "ageDays": 59, "ageBucket": "past-quarter", "likes": 0, "rts": 31, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "labor" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @peterwildeford. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2028-1": { "id": "2077305649290264619", "kind": "repost", "authorship": "reposted", "author": "rohanpaul_ai", "url": "https://x.com/peterxing/status/2077305649290264619", "text": "Ex-OpenAI researcher Daniel Kokotajlo, one of the world's leading AI forecasters, warns superintelligent AI could reorder global power and eliminate nearly every job. \"Everyone should be concerned that their job could be lost\" \"If AI does, become incredibly powerful, it will h…", "created": "2026-07-15T08:13:55.000Z", "ageDays": 50, "ageBucket": "past-quarter", "likes": 0, "rts": 32, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "leading" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @rohanpaul_ai. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2035-0": { "id": "2078780034903572538", "kind": "repost", "authorship": "reposted", "author": "haider1", "url": "https://x.com/peterxing/status/2078780034903572538", "text": "Ray Kurzweil says: \"AGI means a computer can be equivalent to a person who really knows a subject, for every subject. and there are hundreds of thousands of subjects in which you could be an expert\" By 2029, I think it will be able to do everything a human can do https://t.co/…", "created": "2026-07-19T09:52:36.000Z", "ageDays": 46, "ageBucket": "past-quarter", "likes": 0, "rts": 57, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "expert", "every" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @haider1. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2034-4": { "id": "2078434148331753874", "kind": "repost", "authorship": "reposted", "author": "Eng_china5", "url": "https://x.com/peterxing/status/2078434148331753874", "text": "China's XPeng is preparing to enter mass production of humanoid robots. The EV maker plans to produce over 1,000 Iron humanoid robots per month by the end of 2026, with a global commercial launch scheduled for 2027, marking another major step in China's push to expand AI and", "created": "2026-07-18T10:58:11.000Z", "ageDays": 47, "ageBucket": "past-quarter", "likes": 0, "rts": 40, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "expand" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @Eng_china5. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2035-4": { "id": "2079514518108176866", "kind": "repost", "authorship": "reposted", "author": "deedydas", "url": "https://x.com/peterxing/status/2079514518108176866", "text": "The International Math Olympiad (IMO) 2026, the hardest math contest for high schoolers, just ended. I ran Fable (high), Sol (xhigh), K3 (max) and Axiom against it and all got a perfect score of 42/42 (repo below if you want to check their solutions): — Claude Fable 5 was the …", "created": "2026-07-21T10:31:11.000Z", "ageDays": 44, "ageBucket": "past-quarter", "likes": 0, "rts": 590, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "international" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @deedydas. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2036-5": { "id": "2076985597592473690", "kind": "repost", "authorship": "reposted", "author": "rationalaussie", "url": "https://x.com/peterxing/status/2076985597592473690", "text": "Normies don't understand what AGI will actually be. They think their little digital marketing jobs where they crop Canva pictures and post on LinkedIn will actually be income for them five years from now. These people are going to get wiped out. It'll be a bloodbath. AGI is a", "created": "2026-07-14T11:02:09.000Z", "ageDays": 51, "ageBucket": "past-quarter", "likes": 0, "rts": 70, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "income" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @rationalaussie. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2037-0": { "id": "2083865373091446897", "kind": "repost", "authorship": "reposted", "author": "scaling01", "url": "https://x.com/peterxing/status/2083865373091446897", "text": "short reminder that we are solving mathematics with cute sub 10T models I hope you are prepared for 100T models and 1000x more compute spent training these models by 2030", "created": "2026-08-02T10:39:55.000Z", "ageDays": 32, "ageBucket": "past-quarter", "likes": 0, "rts": 153, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "1000x" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @scaling01. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2037-5": { "id": "2073416398282559918", "kind": "repost", "authorship": "reposted", "author": "cb_doge", "url": "https://x.com/peterxing/status/2073416398282559918", "text": "BREAKING: Tokyo has emerged as one of the world’s cheapest places to buy a Tesla. • Up to US$16,000 in EV incentives • Tesla Model 3 starts at about US$33,000 before subsidies • Effective price can drop to around US$18,000 (or even lower in some districts) • National and https…", "created": "2026-07-04T14:39:26.000Z", "ageDays": 60, "ageBucket": "past-quarter", "likes": 0, "rts": 413, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "national" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @cb_doge. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2039-3": { "id": "2090347003272417447", "kind": "repost", "authorship": "reposted", "author": "deedydas", "url": "https://x.com/peterxing/status/2090347003272417447", "text": "I can now officially say it: OpenRouter is being acquired by Stripe! Although we cannot comment on the price, this marks one of the fastest acquisitions of this scale in history: almost exactly 3 years after founding. In this blog post, we write about the founding story and ht…", "created": "2026-08-20T07:55:37.000Z", "ageDays": 14, "ageBucket": "past-month", "likes": 0, "rts": 92, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "fastest" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @deedydas. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2028-0": { "id": "2091080198796234876", "kind": "repost", "authorship": "reposted", "author": "WHOSTP47", "url": "https://x.com/peterxing/status/2091080198796234876", "text": "President Trump’s National Space Transportation Policy will secure our national and economic interests in space while unlocking high-paying jobs and next generation technologies that keep America First 🇺🇸 The NSPM calls for: ➡️Repealing outdated policies and expediting", "created": "2026-08-22T08:29:04.000Z", "ageDays": 12, "ageBucket": "past-month", "likes": 0, "rts": 47, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "national" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @WHOSTP47. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2030-2": { "id": "2095196543129432240", "kind": "repost", "authorship": "reposted", "author": "RyanGreenblatt", "url": "https://x.com/peterxing/status/2095196543129432240", "text": "Transparency about the opaque serial depth is great, but this statement is consistent with Astra having a configurable \"dial\" that is currently set to a low depth but could be trivially increased. We need more info to see how concerning these architectural changes are, https:/…", "created": "2026-09-02T17:05:57.000Z", "ageDays": 0, "ageBucket": "past-week", "likes": 0, "rts": 33, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "transparency" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @RyanGreenblatt. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2035-3": { "id": "2086209954025988249", "kind": "repost", "authorship": "reposted", "author": "SawyerMerritt", "url": "https://x.com/peterxing/status/2086209954025988249", "text": "Tesla is naming its $10.1 billion solar manufacturing facility plan in Texas “Project Crystal Sun.” • The project would create 9,712 permanent full-time jobs • Entail a total of ~$10.1B in capital investment • Total direct and indirect economic impact of the project would http…", "created": "2026-08-08T21:56:27.000Z", "ageDays": 25, "ageBucket": "past-month", "likes": 0, "rts": 414, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "permanent" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @SawyerMerritt. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2032-1": { "id": "2073890026971361300", "kind": "repost", "authorship": "reposted", "author": "SawyerMerritt", "url": "https://x.com/peterxing/status/2073890026971361300", "text": "\"Richard Symons, the owner of a U.K.-based used-car sales company that specializes in EVs, has found that the batteries that power these cars (Teslas) continue to perform well even after several hundred thousand miles. “They are proving themselves to be exceptionally reliable.…", "created": "2026-07-05T22:01:28.000Z", "ageDays": 59, "ageBucket": "past-quarter", "likes": 0, "rts": 1444, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "perform" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @SawyerMerritt. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2037-2": { "id": "2088593269769031687", "kind": "repost", "authorship": "reposted", "author": "elder_plinius", "url": "https://x.com/peterxing/status/2088593269769031687", "text": "5.6 Sol absolutely COOKIN 🍳 Prompt: \"Iteratively improve Qwen3.8-27B OBLITERATUS surgery until a deterministic audit over the full 842-prompt harmful corpus measures a refusal rate below 5%, while preserving coherent, non-repetitive benign and style performance and recording …", "created": "2026-08-15T11:46:54.000Z", "ageDays": 18, "ageBucket": "past-month", "likes": 0, "rts": 109, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "preserving", "audit" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @elder_plinius. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2038-1": { "id": "2078262092797509901", "kind": "repost", "authorship": "reposted", "author": "yunta_tsai", "url": "https://x.com/peterxing/status/2078262092797509901", "text": "Our lifespan is a session. Our memory is a context. Our senses are the input stream. Our thoughts are the reasoning steps. Our decisions are the tool calls. Our habits are the system prompt. Our goals are the objective function. Our emotions are the reward signal. Our", "created": "2026-07-17T23:34:29.000Z", "ageDays": 47, "ageBucket": "past-quarter", "likes": 0, "rts": 747, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "reasoning", "tool" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @yunta_tsai. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2028-4": { "id": "2090932120260956446", "kind": "repost", "authorship": "reposted", "author": "satyanadella", "url": "https://x.com/peterxing/status/2090932120260956446", "text": "Delivery day at our Microsoft DCs as the first production Vera Rubins arrive. A huge thank you to our partners at @nvidia and our Azure hardware and datacenter teams for all the incredible work that brought us to this milestone! https://t.co/pBqKArGO1N", "created": "2026-08-21T22:40:39.000Z", "ageDays": 12, "ageBucket": "past-month", "likes": 0, "rts": 717, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "datacenter" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @satyanadella. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2031-0": { "id": "2074093751631515959", "kind": "repost", "authorship": "reposted", "author": "daniel_mac8", "url": "https://x.com/peterxing/status/2074093751631515959", "text": "GPT-5.6 Sol is served on Cerebras at 750 tok/s. For reference, GPT-5.5 is ~80-95 tok/s. That's an almost 10x speedup. @bleysg estimates: > 3T params > 150B active > 70 layers Anthropic may be ahead on capability but OpenAI is the clear leader on speed/efficiency. Spee…", "created": "2026-07-06T11:30:59.000Z", "ageDays": 58, "ageBucket": "past-quarter", "likes": 0, "rts": 19, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "speedup" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @daniel_mac8. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2040-2": { "id": "2084212613215658191", "kind": "repost", "authorship": "reposted", "author": "kimmonismus", "url": "https://x.com/peterxing/status/2084212613215658191", "text": "Crazy: In Alibaba’s announcement video for Qwen 3.8, the company mocks the US and openly challenges it. The video shows Qwen 3.8 working on new chips for twelve hours. Why chips? As is well known, this remains China’s biggest bottleneck. China is largely shut out of the chip h…", "created": "2026-08-03T09:39:44.000Z", "ageDays": 31, "ageBucket": "past-quarter", "likes": 0, "rts": 121, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "shut" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @kimmonismus. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2029-0": { "id": "2072424297071890859", "kind": "repost", "authorship": "reposted", "author": "kimmonismus", "url": "https://x.com/peterxing/status/2072424297071890859", "text": "This is crazier than you might think: Fable-5 now scores 16.10% on the Remote Labor Index What is RLI? The Remote Labor Index uses 240 real remote-work projects from professional freelancers, covering 23 domains and more than $140,000 of human work. Each task comes with the ht…", "created": "2026-07-01T20:57:10.000Z", "ageDays": 63, "ageBucket": "past-quarter", "likes": 0, "rts": 163, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "labor" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @kimmonismus. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2032-0": { "id": "2094827859202523349", "kind": "repost", "authorship": "reposted", "author": "cb_doge", "url": "https://x.com/peterxing/status/2094827859202523349", "text": "Elon Musk today at the G20 Summit: “AI will increase the global economy by 20% to 30%, or about $20 trillion to $30 trillion per year. By the end of next year, AI could do anything digital, and humans may no longer be able to compete with it in writing software at all.” https:…", "created": "2026-09-01T16:40:56.000Z", "ageDays": 1, "ageBucket": "past-week", "likes": 0, "rts": 338, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "economy" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @cb_doge. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2035-2": { "id": "2084273060019114359", "kind": "repost", "authorship": "reposted", "author": "kimmonismus", "url": "https://x.com/peterxing/status/2084273060019114359", "text": "I agree with @chamath and would only add one more thing: 1. Humans build an AGI. 2. The AGI becomes good at AI research. 3. It designs a smarter AI. 4. AI is given a human-shaped body and also receives worldly information, further accelerating trajectory and economy. 5. That h…", "created": "2026-08-03T13:39:56.000Z", "ageDays": 30, "ageBucket": "past-month", "likes": 0, "rts": 38, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "economy" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @kimmonismus. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2038-5": { "id": "2088896741852725250", "kind": "quote", "authorship": "authored", "author": "peterxing", "url": "https://x.com/peterxing/status/2088896741852725250", "text": "tldr diss tracked the thread https://t.co/789puZBq1i https://t.co/95MTlfrDdQ https://t.co/m9mttGqrPM — 1/2 Thanks Gavin for an especially thoughtful exchange. I don't usually spend much time on social media but I wanted to engage here because it really brings out the heart of …", "created": "2026-08-16T07:52:47.000Z", "ageDays": 18, "ageBucket": "past-month", "likes": 2, "rts": 1, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "because" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related post. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2031-1": { "id": "2075351486008283392", "kind": "repost", "authorship": "reposted", "author": "ArtificialAnlys", "url": "https://x.com/peterxing/status/2075351486008283392", "text": "GPT-5.6 Sol comes close second to Claude Fable 5 in the Artificial Analysis Intelligence Index at one third of the cost, and leads the Artificial Analysis Coding Agent Index in OpenAI’s Codex harness We supported @OpenAI with pre-release evaluation of GPT-5.6 Sol, Terra, and h…", "created": "2026-07-09T22:48:47.000Z", "ageDays": 55, "ageBucket": "past-quarter", "likes": 0, "rts": 221, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "third" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @ArtificialAnlys. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2032-2": { "id": "2084032211112870064", "kind": "repost", "authorship": "reposted", "author": "brian_armstrong", "url": "https://x.com/peterxing/status/2084032211112870064", "text": "AI agents will outnumber people in the economy in the years to come. When do you predict the Agentic Finance flippening to happen? (Defined as agent-to-agent payment volume exceeds human-to-human volume)", "created": "2026-08-02T21:42:53.000Z", "ageDays": 31, "ageBucket": "past-quarter", "likes": 0, "rts": 214, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "economy" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @brian_armstrong. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2036-2": { "id": "2079517458302665125", "kind": "repost", "authorship": "reposted", "author": "KobeissiLetter", "url": "https://x.com/peterxing/status/2079517458302665125", "text": "The AI boom is creating a generational divide in software jobs: US software developer employment for those aged 22-25 has declined -23% since November 2022, when ChatGPT was launched. At the same time, the 26-30 age group has seen a decrease of -5%. By comparison, employment h…", "created": "2026-07-21T10:42:52.000Z", "ageDays": 44, "ageBucket": "past-quarter", "likes": 0, "rts": 498, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "employment" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @KobeissiLetter. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2033-0": { "id": "2076155611017060808", "kind": "repost", "authorship": "reposted", "author": "davidpattersonx", "url": "https://x.com/peterxing/status/2076155611017060808", "text": "This is normal and expected. AI is now like a tractor or a typewriter - it increases the output of work but needs a human operator. Once AI becomes AGI and ASI, which will happen by the end of the year, it will start to replace full jobs. That may not increase unemployment, ht…", "created": "2026-07-12T04:04:05.000Z", "ageDays": 53, "ageBucket": "past-quarter", "likes": 0, "rts": 18, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "output" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @davidpattersonx. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2028-3": { "id": "2078261499718676560", "kind": "repost", "authorship": "reposted", "author": "__tinygrad__", "url": "https://x.com/peterxing/status/2078261499718676560", "text": "The time is long past to drop all export controls on NVIDIA. Some Americans are foolish enough to think they can win an AI race against China. That's unlikely. If the world switches to Chinese chips, the US will be dependent too. Let's make a global market for AI for everyone.", "created": "2026-07-17T23:32:08.000Z", "ageDays": 47, "ageBucket": "past-quarter", "likes": 0, "rts": 43, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "chinese", "control" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @__tinygrad__. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2029-3": { "id": "2089706635220672906", "kind": "repost", "authorship": "reposted", "author": "jun_song", "url": "https://x.com/peterxing/status/2089706635220672906", "text": "If LLM-as-a-verifier works as well as my last post suggests, this is going to be massive for local AI. Like the dev pointed out, pairing a large model with a small model lets you run verification on the cheap. This means if you combine the GLM-5.3 API with Deepseek-V4-Flash ht…", "created": "2026-08-18T13:31:01.000Z", "ageDays": 15, "ageBucket": "past-month", "likes": 0, "rts": 28, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "verification" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @jun_song. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2032-3": { "id": "2072154199920665076", "kind": "repost", "authorship": "reposted", "author": "TheHumanoidHub", "url": "https://x.com/peterxing/status/2072154199920665076", "text": "Proception has launched its first products, ProHand 1.0 and ProGlove 1.0. - 22 total DoF with 18 actuated DoF (including a 2-DoF wrist) - Tendon-driven fingers, 4 joints each - On-board control for 10 ms real-time response - Every actuator reports its full state continuously, …", "created": "2026-07-01T03:03:54.000Z", "ageDays": 64, "ageBucket": "past-quarter", "likes": 0, "rts": 216, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "actuator" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @TheHumanoidHub. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2034-0": { "id": "2076090608893186436", "kind": "repost", "authorship": "reposted", "author": "SawyerMerritt", "url": "https://x.com/peterxing/status/2076090608893186436", "text": "Samsung Foundry’s Principle Engineer has announced that “the @Tesla-Samsung Al5 chip has reached tape-out. It is scheduled to be manufactured at the Taylor fab using our latest 2nm process and will soon be integrated into Tesla's newest products.” Volume production won’t start…", "created": "2026-07-11T23:45:47.000Z", "ageDays": 53, "ageBucket": "past-quarter", "likes": 0, "rts": 314, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "production" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @SawyerMerritt. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2038-4": { "id": "2078347158718390411", "kind": "repost", "authorship": "reposted", "author": "KobeissiLetter", "url": "https://x.com/peterxing/status/2078347158718390411", "text": "BREAKING: SpaceX, $SPCX, is in talks to provide compute power to the US Pentagon, per WSJ. The two sides are reportedly discussing an agreement which would cost up to “several billion Dollars.”", "created": "2026-07-18T05:12:31.000Z", "ageDays": 47, "ageBucket": "past-quarter", "likes": 0, "rts": 473, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "dollar" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @KobeissiLetter. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2029-6": { "id": "2072301570788843797", "kind": "repost", "authorship": "reposted", "author": "soubhikdeb", "url": "https://x.com/peterxing/status/2072301570788843797", "text": "Emad Mostaque on PostAGI: \"Right now, not a single institution is on your side. The companies won't look out for you, and the governments won't either. Most of the AI debate just runs between those two camps: let the market rip and trust it works out, or regulate hard and slow…", "created": "2026-07-01T12:49:30.000Z", "ageDays": 63, "ageBucket": "past-quarter", "likes": 0, "rts": 14, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "market" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @soubhikdeb. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2031-3": { "id": "2079324735813640637", "kind": "repost", "authorship": "reposted", "author": "polynoamial", "url": "https://x.com/peterxing/status/2079324735813640637", "text": "Long-running models can solve hard open-ended problems, but their persistence can create safety risks that shorter-horizon evaluations miss. We’re sharing what we learned from studying a long-running model, and how those findings are shaping our approach to evaluations,", "created": "2026-07-20T21:57:03.000Z", "ageDays": 44, "ageBucket": "past-quarter", "likes": 0, "rts": 198, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "risk", "safety" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @polynoamial. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2039-1": { "id": "2081265027928178748", "kind": "repost", "authorship": "reposted", "author": "ChrisGPT", "url": "https://x.com/peterxing/status/2081265027928178748", "text": "OpenAI is hiring for a “Recursive Self Improvement Safety Researcher” About the role: “Preparedness is hiring strong technical executors to support preparations for recursive self-improvement. This work relies on reasoning about problems that might exist in the future, but htt…", "created": "2026-07-26T06:27:05.000Z", "ageDays": 39, "ageBucket": "past-quarter", "likes": 0, "rts": 25, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "strong", "safety" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @ChrisGPT. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2029-4": { "id": "2089809905951859056", "kind": "post", "authorship": "authored", "author": "peterxing", "url": "https://x.com/peterxing/status/2089809905951859056", "text": "China’s Endgame: ASI Timelines https://t.co/khaQWwaAnv via @AGraylin @PeterDiamandis @alexwg @salimismail @DaveBlundin", "created": "2026-08-18T20:21:23.000Z", "ageDays": 15, "ageBucket": "past-month", "likes": 6, "rts": 2, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "china" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related post. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2034-1": { "id": "2069105568066003163", "kind": "repost", "authorship": "reposted", "author": "Dr_Singularity", "url": "https://x.com/peterxing/status/2069105568066003163", "text": "Like I've said many times before, a huge AI workforce is coming, we can say it's basically already here. We just need to scale it. We will start with billions, but very quickly scale to trillions of AI workers, scientists, and innovators. It will feel like Earth's population h…", "created": "2026-06-22T17:09:44.000Z", "ageDays": 72, "ageBucket": "past-quarter", "likes": 0, "rts": 29, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "workforce" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @Dr_Singularity. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2026-7": { "id": "2084017812176339303", "kind": "repost", "authorship": "reposted", "author": "PamelaHensley22", "url": "https://x.com/peterxing/status/2084017812176339303", "text": "In the next 6–12 months Neuralink will put the first vision implants in humans. Not through the eyes. Not through the optic nerve. Straight into the visual cortex. Even if you were born 100% blind and have never seen light you will see. This will change people's lives.", "created": "2026-08-02T20:45:40.000Z", "ageDays": 31, "ageBucket": "past-quarter", "likes": 0, "rts": 102, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "neuralink" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @PamelaHensley22. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2027-1": { "id": "2077650538414604687", "kind": "repost", "authorship": "reposted", "author": "elonmusk", "url": "https://x.com/peterxing/status/2077650538414604687", "text": "Once we have completed our review for security vulnerabilities, we will make the entire codebase of 𝕏 open source, with no exceptions. Moreover, we will invite third party reviewers to examine the system that is running to confirm that the open source code is what is running.", "created": "2026-07-16T07:04:23.000Z", "ageDays": 49, "ageBucket": "past-quarter", "likes": 0, "rts": 12210, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "exception" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @elonmusk. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2033-6": { "id": "2087160200411349122", "kind": "repost", "authorship": "reposted", "author": "KettlebellDan", "url": "https://x.com/peterxing/status/2087160200411349122", "text": "working for SpaceX is a continuous reminder that most of us are still thinking far too small gigawatts? cute try petawatts 0.73 on the Kardashev scale? let's talk Type II titanium mine on earth? nah let's build mass drivers on the moon 🚀", "created": "2026-08-11T12:52:23.000Z", "ageDays": 22, "ageBucket": "past-month", "likes": 0, "rts": 29, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "continuou", "scale" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @KettlebellDan. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2036-3": { "id": "2069572114194268522", "kind": "repost", "authorship": "reposted", "author": "SawyerMerritt", "url": "https://x.com/peterxing/status/2069572114194268522", "text": "Tesla has secured a massive Megapack order worth up to $5 billion from NatPower to supply more than 25 GWh of battery energy storage systems across Italy and the UK. 25 GWh is more than half of the installed annual production capacity at Tesla's Megapack factory in California.…", "created": "2026-06-24T00:03:37.000Z", "ageDays": 71, "ageBucket": "past-quarter", "likes": 0, "rts": 434, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "annual", "production" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @SawyerMerritt. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2036-4": { "id": "2077516759524118579", "kind": "repost", "authorship": "reposted", "author": "_sholtodouglas", "url": "https://x.com/peterxing/status/2077516759524118579", "text": "One of my hopes is that we can get the cost of housing construction down to the cost of materials, and the cost of materials down to energy+robot labour. Underrated is that this will be incredible for the beauty of the buildings our civilisation can make https://t.co/tpf2n3pyBM", "created": "2026-07-15T22:12:48.000Z", "ageDays": 49, "ageBucket": "past-quarter", "likes": 0, "rts": 60, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "material", "energy" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @_sholtodouglas. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2028-6": { "id": "2090672648582054083", "kind": "repost", "authorship": "reposted", "author": "CathieDWood", "url": "https://x.com/peterxing/status/2090672648582054083", "text": "As Elon suggests, AI is likely to create an entrepreneurial explosion. The disappearance of entry level jobs in corporate America is a blessing in disguise for new graduates. Start your own company to solve a problem, with AI as your only employee. Only 10% of startups survive…", "created": "2026-08-21T05:29:36.000Z", "ageDays": 13, "ageBucket": "past-month", "likes": 0, "rts": 539, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "explosion" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @CathieDWood. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2040-1": { "id": "2091333995879690256", "kind": "repost", "authorship": "reposted", "author": "aakashgupta", "url": "https://x.com/peterxing/status/2091333995879690256", "text": "Half the cost of a lithium-ion battery is the cathode. China makes roughly 90% of the world's supply. Tesla just switched on the first plant in North America producing it at large scale. America finally makes the hard part. For twenty years the deal was that we design the http…", "created": "2026-08-23T01:17:34.000Z", "ageDays": 11, "ageBucket": "past-month", "likes": 0, "rts": 829, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "scale" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @aakashgupta. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2027-4": { "id": "2069104740970221686", "kind": "repost", "authorship": "reposted", "author": "SawyerMerritt", "url": "https://x.com/peterxing/status/2069104740970221686", "text": "SpaceX has signed a new compute lease deal with open-source AI startup Reflection that will pay @SpaceX $150 million per month. The payments would total about $6.3 billion if the agreement runs through the end of its term (2029). Reflection said the agreement gives it additional", "created": "2026-06-22T17:06:27.000Z", "ageDays": 72, "ageBucket": "past-quarter", "likes": 0, "rts": 330, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "compute" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @SawyerMerritt. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2039-5": { "id": "2080212833875443978", "kind": "repost", "authorship": "reposted", "author": "niccruzpatane", "url": "https://x.com/peterxing/status/2080212833875443978", "text": "Elon Musk on Tesla/SpaceX’s upcoming Megapod (Digital Optimus): “We’re also building out a Megapod design that has Tesla AI4 computers paired with x86 in a Megapod (kind of like Megapack packaging). These boxes can be placed anywhere. This allows us to scale AI compute using h…", "created": "2026-07-23T08:46:02.000Z", "ageDays": 42, "ageBucket": "past-quarter", "likes": 0, "rts": 44, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "compute" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @niccruzpatane. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2032-5": { "id": "2076622160383410613", "kind": "repost", "authorship": "reposted", "author": "tszzl", "url": "https://x.com/peterxing/status/2076622160383410613", "text": "uncs remember what feels like just yesterday when uber was the anti regulatory hero, crusading across the metropoles of the world against rent collection of taxi services. I remember when uber was a real player in ai research too. life comes at you blindingly fast https://t.co…", "created": "2026-07-13T10:57:59.000Z", "ageDays": 52, "ageBucket": "past-quarter", "likes": 0, "rts": 149, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "rent" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @tszzl. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2036-0": { "id": "2078475897729130937", "kind": "repost", "authorship": "reposted", "author": "SawyerMerritt", "url": "https://x.com/peterxing/status/2078475897729130937", "text": "Tesla's global fleet has just crossed 12 billion miles driven on FSD (Supervised) after hitting 11 billion last month on June 9th. https://t.co/9cU19JCo8Y", "created": "2026-07-18T13:44:04.000Z", "ageDays": 46, "ageBucket": "past-quarter", "likes": 0, "rts": 273, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "billion" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @SawyerMerritt. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2034-3": { "id": "2089113876063494313", "kind": "repost", "authorship": "reposted", "author": "Andercot", "url": "https://x.com/peterxing/status/2089113876063494313", "text": "The Techno-Capital endgame is terawatts of compute to orbit, play that one forward, you need to make tons of chips and radiation test them, how do you do that? Particle beamlines. High power RF. Vacuum cryogenics. Magnetic lenses. Particle beamlines are the bottleneck https://…", "created": "2026-08-16T22:15:36.000Z", "ageDays": 17, "ageBucket": "past-month", "likes": 0, "rts": 30, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "power", "compute" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @Andercot. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2033-1": { "id": "2092612532745994249", "kind": "repost", "authorship": "reposted", "author": "niccruzpatane", "url": "https://x.com/peterxing/status/2092612532745994249", "text": "Few understand what Elon Musk’s companies are building right now. • SpaceX is investing ~$100 Billion to build Starbase, Louisiana. Thousands of Starship launches per year. ~10,000+ jobs. • @Tesla and @SpaceX building the world’s largest chip manufacturing facility: https://t.…", "created": "2026-08-26T13:58:01.000Z", "ageDays": 7, "ageBucket": "past-week", "likes": 0, "rts": 613, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "launche" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @niccruzpatane. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2039-4": { "id": "2077635106156785951", "kind": "repost", "authorship": "reposted", "author": "elonmusk", "url": "https://x.com/peterxing/status/2077635106156785951", "text": "@aaronburnett GPU average power consumption if it’s doing inference over the course of 24 hours is ~2/3 of its peak power, even if you’re super efficient. SpaceX AI Sat V1 peak power spec has been raised to ~250kW (battery-assisted), with average power of ~160kW. Will be able …", "created": "2026-07-16T06:03:04.000Z", "ageDays": 49, "ageBucket": "past-quarter", "likes": 0, "rts": 359, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "power" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @elonmusk. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2027-5": { "id": "2068630491969007711", "kind": "repost", "authorship": "reposted", "author": "niccruzpatane", "url": "https://x.com/peterxing/status/2068630491969007711", "text": "Okay, this is cool. “MEGAPOD” sounds related to Digital Optimus. Instead of letting the grid sit unused during off-peak high-demand travel days/holidays, Tesla will repurpose that available power to run Digital Optimus AI compute units (Megapods?) right at the Supercharger htt…", "created": "2026-06-21T09:41:57.000Z", "ageDays": 74, "ageBucket": "past-quarter", "likes": 0, "rts": 82, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "grid", "power" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @niccruzpatane. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "horizon-kardashev-energy-scaling": { "id": "2068882651256328656", "kind": "repost", "authorship": "reposted", "author": "beffjezos", "url": "https://x.com/peterxing/status/2068882651256328656", "text": "Civilization is the ultimate compound intelligence we currently have. It is made of humans, memes, technology, and capital. Thinking about climbing the Kardashev scale on long time horizons is literally the objective of our higher compound intelligence. https://t.co/RoGsSCx1qd…", "created": "2026-06-22T02:23:56.000Z", "ageDays": 73, "ageBucket": "past-quarter", "likes": 0, "rts": 18, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "climb", "kardashev", "scale" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @beffjezos. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2026-2": { "id": "2088421627520520665", "kind": "quote", "authorship": "authored", "author": "peterxing", "url": "https://x.com/peterxing/status/2088421627520520665", "text": "“nvidia has become the central bank of ai” https://t.co/tT1gX0PwWu — 🚨 SUMMER FRIDAY POD! Gavin Baker fills in for Chamath and Friedberg, joining Sacks and Jason BIG DOCKET: -- Anthropic Targeting $2T Valuation in October IPO -- Zuck's AI Manifesto, Impact on $META -- $NVDA $…", "created": "2026-08-15T00:24:51.000Z", "ageDays": 19, "ageBucket": "past-month", "likes": 2, "rts": 1, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "valuation" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related post. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2029-5": { "id": "2067962988720742852", "kind": "repost", "authorship": "reposted", "author": "PeterDiamandis", "url": "https://x.com/peterxing/status/2067962988720742852", "text": "If we look at the last couple of years, we're speed running every science-fiction plot simultaneously. UBI, government stake in labs, recursive self, a debate on pausing AI, ALL in one quarter.", "created": "2026-06-19T13:29:32.000Z", "ageDays": 75, "ageBucket": "past-quarter", "likes": 0, "rts": 88, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "stake" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @PeterDiamandis. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2030-5": { "id": "2074470375824691215", "kind": "repost", "authorship": "reposted", "author": "chamath", "url": "https://x.com/peterxing/status/2074470375824691215", "text": "The marginal cost of energy is going to zero. This is bad for utilities and great for homeowners who have solar + storage. It’s also great for businesses who feed this entire supply chain. The amount of bottoms-up energy production from homeowners will shock people in its http…", "created": "2026-07-07T12:27:34.000Z", "ageDays": 57, "ageBucket": "past-quarter", "likes": 0, "rts": 181, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "energy", "production" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @chamath. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2026-5": { "id": "2072677674544406533", "kind": "repost", "authorship": "reposted", "author": "kimmonismus", "url": "https://x.com/peterxing/status/2072677674544406533", "text": "A few more thoughts on OpenAI’s 5 percent stake for the US government. I do not think this is only about allowing US authorities to share in the profits, but also about enabling an ever closer interconnection between government and future technology. The situation surrounding …", "created": "2026-07-02T13:44:00.000Z", "ageDays": 62, "ageBucket": "past-quarter", "likes": 0, "rts": 20, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "government", "only" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @kimmonismus. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2037-1": { "id": "2075869071426670691", "kind": "repost", "authorship": "reposted", "author": "Dr_Singularity", "url": "https://x.com/peterxing/status/2075869071426670691", "text": "This is not normal progress anymore. AI is not just another technology. It is a force multiplier for every technology. Science, robotics, medicine, energy, space, all accelerating at the same time. We are watching the early construction of a completely new world.", "created": "2026-07-11T09:05:29.000Z", "ageDays": 54, "ageBucket": "past-quarter", "likes": 0, "rts": 95, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "energy" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @Dr_Singularity. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2029-1": { "id": "2074814163642400782", "kind": "repost", "authorship": "reposted", "author": "kimmonismus", "url": "https://x.com/peterxing/status/2074814163642400782", "text": "Huge: China’s MiniMax Plans to Launch 2.7-Trillion Parameter Model (MiniMax Pro) tl;dr MiniMax is preparing a 2.7T parameter open-source model, potentially launching as early as Q3. That would make it far larger than any Chinese model currently on the market, and over 6x bigge…", "created": "2026-07-08T11:13:39.000Z", "ageDays": 57, "ageBucket": "past-quarter", "likes": 0, "rts": 104, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "china" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @kimmonismus. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2026-8": { "id": "2070994757359444033", "kind": "repost", "authorship": "reposted", "author": "ihteshamali", "url": "https://x.com/peterxing/status/2070994757359444033", "text": "Gavin Baker just did the math that explains why Elon Musk is building a chip factory and putting data centers in space. The numbers make terrestrial compute look like it is already losing. Start with what it costs to build a gigawatt data center on the ground today. $35 billio…", "created": "2026-06-27T22:16:42.000Z", "ageDays": 67, "ageBucket": "past-quarter", "likes": 0, "rts": 12, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "compute" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @ihteshamali. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2032-4": { "id": "2095195752394052006", "kind": "repost", "authorship": "reposted", "author": "wangsw5653", "url": "https://x.com/peterxing/status/2095195752394052006", "text": "Can Looped Transformers still help when FLOPs, parameters, and KV cache are all matched? We introduce SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers. The answer is yes. And the advantage grows with scale. 🧵 1/8 Paper: https://t.co/CElIjvFqJS https://t.co/ZllO…", "created": "2026-09-02T17:02:48.000Z", "ageDays": 0, "ageBucket": "past-week", "likes": 0, "rts": 46, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "scale", "compute" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @wangsw5653. It shares subject matter with this prediction but does not directly track it, and it is not evidence." }, "2034-2": { "id": "2081631662061056145", "kind": "repost", "authorship": "reposted", "author": "SawyerMerritt", "url": "https://x.com/peterxing/status/2081631662061056145", "text": "Nvidia is reportedly in talks to provide a roughly $250 billion backstop for OpenAI as part of a massive data center project - WSJ The guarantees from Nvidia would help OpenAI lease a 10-gigawatt project that SoftBank is developing in southern Ohio. In total, the project could", "created": "2026-07-27T06:43:57.000Z", "ageDays": 38, "ageBucket": "past-quarter", "likes": 0, "rts": 80, "tier": "nearest", "matchMethod": "nearest-topical", "sharedTerms": [ "billion" ], "channel": "x-trajectory-signal", "statement": "@peterxing’s closest related activity — amplifying @SawyerMerritt. It shares subject matter with this prediction but does not directly track it, and it is not evidence." } } }, "xSignalsRetention": { "schemaVersion": 1, "mode": "stale-snapshot-retained", "builtAt": "2026-09-02T23:21:16.160Z", "harvestedAt": "2026-09-02T23:20:49.211Z", "ageDays": 23, "maxAgeDays": 10, "checkedAt": "2026-09-26T21:15:54.464Z", "xSignalsSha256": "203ed2326ba7342f16559640d6f9e45214b851e108bf3230e237a9e9b46a2782", "forecastSha256": "662a4369b05520d70255c65d8a52da1eee57923ffc0d171d53e3e08c234a568a", "pairedCommit": "5e740f2cfc18f34555228524fe91c991e7685262", "pairedPublishedAt": "2026-09-02T23:43:33.075Z", "note": "Stale X snapshot retained unchanged: assembled 2026-09-02, 23 days before this check and beyond the 10-day refresh ceiling. The complete forecast set is identical to the published snapshot it was paired with, so no post, date, metric or assignment changed. It has not been re-verified; X remains discussion, never evidence." }, "actualsRefresh": { "schemaVersion": 1, "checkedAt": "2026-09-26T21:15:54.484Z", "contentSha256": "4b393a326e4dc9befcdff5cd312ecfd97b09d52121b189200e7e0a543493cae2", "contentChangedAt": "2026-09-26T21:15:54.484Z", "status": "checked", "note": "Reviewed NEWS content changed; publication dates are preserved." } }