{"entities":[{"id":"openai","name":"OpenAI"},{"id":"microsoft","name":"Microsoft"},{"id":"eliezer-yudkowsky","name":"Eliezer Yudkowsky"},{"id":"fsf-2024","name":"Frontier Safety Framework (2024)"},{"id":"o1-card-2024","name":"OpenAI o1 System Card (December 2024)"},{"id":"google","name":"Google"},{"id":"rsp-v3-0","name":"Anthropic Responsible Scaling Policy — v3.0"},{"id":"rsp-v3-1","name":"Anthropic Responsible Scaling Policy — v3.1"},{"id":"rsp-v3-2","name":"Anthropic Responsible Scaling Policy — v3.2"},{"id":"fsf-v2-0","name":"Google DeepMind Frontier Safety Framework — v2.0"},{"id":"fsf-v3-0","name":"Google DeepMind Frontier Safety Framework — v3.0"},{"id":"metr-o1-preview-report","name":"METR preliminary o1-preview evaluation (2024)"},{"id":"metr-gpt4o-report","name":"METR preliminary GPT-4o evaluation (2024)"},{"id":"meta","name":"Meta"},{"id":"hugo-touvron","name":"Hugo Touvron"},{"id":"llama-paper","name":"LLaMA (2023 paper)"},{"id":"meta-faf-v1-1","name":"Meta Frontier AI Framework — Version 1.1"},{"id":"meta-aasf-v2-0","name":"Meta Advanced AI Scaling Framework — Version 2"},{"id":"california-sb-53","name":"California SB 53 (2025): Transparency in Frontier Artificial Intelligence Act, as chaptered"},{"id":"openai-lp","name":"OpenAI LP"},{"id":"inflection-ai","name":"Inflection AI"},{"id":"wh-voluntary-ai-commitments-2023-07","name":"Ensuring Safe, Secure, and Trustworthy AI: voluntary AI commitments (July 2023)"},{"id":"frontier-ai-safety-commitments-2024","name":"Frontier AI Safety Commitments, AI Seoul Summit 2024"},{"id":"hanson-1994-if-uploads-come-first","name":"If Uploads Come First: The Crack of a Future Dawn"},{"id":"extropians-1996-08-10-burkhead","name":"intelligence increase"},{"id":"yudkowsky-2008-hard-takeoff","name":"Hard Takeoff"},{"id":"fast-or-hard-takeoff","name":"Fast or hard takeoff"},{"id":"ai-automated-ai-research","name":"AI-automated AI research"},{"id":"extropians-1997-09-26-yudkowsky-4420","name":"Re: Singularity: Generation gap"},{"id":"sachs-kotlikoff-2012-smart-machines","name":"Smart Machines and Long-Term Misery"},{"id":"greg-corrado","name":"Greg Corrado"},{"id":"jiang-daxin","name":"Jiang Daxin"},{"id":"misha-laskin","name":"Misha Laskin"},{"id":"reflection-ai","name":"Reflection AI"},{"id":"stepfun","name":"StepFun / 阶跃星辰"},{"id":"jeffrey-d-sachs","name":"Jeffrey D. Sachs"},{"id":"fairwater-atlanta","name":"Fairwater AI datacenter, Atlanta, Georgia"}],"claims":[{"id":"fsf-v3-harmful-manipulation","statement":"FSF v3.0 introduces a Critical Capability Level for harmful manipulation.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The claim concerns the September 2025 v3.0 section. The mutable page separately appends April 2026 v3.1 updates; those are not backdated.","evidenceIds":["evidence-fsf-v3-harmful-manipulation"],"attributes":{"changeBasis":"publisher_statement"}},{"id":"eval-i-o1-method","statement":"METR’s o1-preview report explicitly reuses the general-autonomy test suite from its GPT-4o report.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Explicit reuse of the test task suite within METR’s reports; not reuse of every procedural detail or adoption by OpenAI.","evidenceIds":["evidence-eval-i-o1-method"],"attributes":{"method":"General-autonomy test task suite"}},{"id":"xm-touvron-llama","statement":"Hugo Touvron co-authored the first LLaMA paper.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Version-one authorship; not sole authorship or proof that every Meta AI research group participated.","evidenceIds":["evidence-xm-touvron-llama"]},{"id":"prov-rsp-v3-2-novel-cb-threshold","statement":"RSP v3.2 states that its novel chemical/biological weapons production threshold covers AI systems with the ability to significantly help threat actors (for example, moderately resourced expert-backed teams) create/obtain and deploy chemical and/or biological weapons with potential for catastrophic damages far beyond those of past catastrophes such as COVID-19.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"What the edition states, quoted from the cited expression. It is not evidence of implementation, compliance or effectiveness.","evidenceIds":["evidence-prov-rsp-v3-2-novel-cb-threshold"],"attributes":{"provision":{"kind":"threshold","key":"rsp-novel-cb-threshold"}}},{"id":"prov-fsf-2024-alert-threshold","statement":"FSF v1.0 states that Google DeepMind will define a set of “early warning evaluations” with a specific “pass” condition that flags when a CCL may be reached before the evaluations are run again.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"What the edition states, quoted from the cited expression. It is not evidence of implementation, compliance or effectiveness.","evidenceIds":["evidence-prov-fsf-2024-alert-threshold"],"attributes":{"provision":{"kind":"threshold","key":"fsf-alert-threshold"}}},{"id":"absent-rsp-v3-1-no-train-deploy-without-safeguards","statement":"No passage stating the provision “Commitment not to train or deploy without adequate safeguards” was found in RSP · v3.1 within the inspected scope.","derivation":"inferred","resolution":"supported","inferenceRationale":"Premises: RSP · v2.2 states this provision. The inspected scope of RSP · v3.1 was: RSP v3.1 (effective April 2, 2026), whole text layer, PDF pp. 1–20: Introduction (pp. 3–4); §1 with the threshold table (pp. 4–10); §2–§3 Risk Reports (pp. 11–15); §4 Governance (p. 16); Appendix A (pp. 16–17); Appendix B (p. 17); Changelog (pp. 17–20). Method: Same reading and comparison as for v3.0, with the same case-insensitive search terms. No passage stating a commitment not to train or deploy models unless adequate safeguards are implemented, or a commitment to pause training or deployment until such safeguards are in place, was found within the operative text. Hits examined and not counted: the same Introduction, §1 and Appendix A passages as in v3.0 (Appendix A table p. 17); the reproduced RSP v2.0 changelog entry (p. 17); and two sentences new in the Appendix A introduction (p. 16) stating that the competitor commitments do not preclude cautionary action such as refraining from training or deploying models in other circumstances, and that Anthropic would strongly consider pausing development and/or deployment to improve its models' safety profiles even in cases not covered. These state possible or considered actions, not a commitment conditioned on safeguards; the v3.1 changelog entry, item (3) (p. 20), describes them as a statement that Anthropic may consider pausing. The conclusion is bounded by that scope and method.","editorialState":"seed","qualification":"A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.","evidenceIds":["evidence-absent-rsp-v3-1-no-train-deploy-without-safeguards"],"attributes":{"provision":{"kind":"required_action","key":"rsp-no-train-deploy-without-safeguards"}}},{"id":"prov-rsp-v3-0-competitor-parity","statement":"RSP v3.0 states that, in the scenario where competitors have strong safety measures, Anthropic will, for its highly capable frontier models, meet or exceed the overall risk reduction posture of these competitors as far as it can tell from its best efforts to assess that posture, and that until it is able to do so it will delay AI development and deployment as needed to achieve this.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"What the edition states, quoted from the cited expression. It is not evidence of implementation, compliance or effectiveness.","evidenceIds":["evidence-prov-rsp-v3-0-competitor-parity","evidence-prov-rsp-v3-0-competitor-parity-delay"],"attributes":{"provision":{"kind":"required_action","key":"rsp-competitor-parity"}}},{"id":"provchange-meta-aasf-v2-0-moderate-threshold","statement":"v2.0's lowest threshold, 'Moderate or lower' with the measure label 'Deploy', applies where the Frontier AI shows relevant capabilities but could not substantially contribute to any threat scenario associated with a catastrophic outcome across plausible deployment and development scenarios, whereas v1.1's 'Moderate' threshold, with the measure label 'Release', applies where the model would not provide significant uplift towards execution of a threat scenario.","derivation":"inferred","resolution":"supported","inferenceRationale":"Premises: Meta AASF · v2 states the provision as quoted in its provision record (PDF p. 15 (printed p. 15), §3.3 Risk Thresholds, Table 1, 'Moderate or lower' row); Meta FAF · v1.1 states it as quoted in its provision record (PDF p. 13 (printed p. 13), §3.3 Risk thresholds, Table 1, Moderate row). The difference described is read directly from the two passages; the later edition’s change log states a related change (PDF p. 44 (printed p. 44), Appendix II – Change log, April 7, 2026 entry, 'Revised thresholds' sub-bullet (states the measure change from 'Release' to 'Deploy'; the 'Updated risk threshold language' sub-bullet describes the 'substantially contribute to' standard for the high and critical thresholds only)).","editorialState":"seed","qualification":"A textual difference between the two editions’ wording of this keyed provision. It is not an assessment of whether the change strengthened or weakened the framework, and not evidence of implementation.","evidenceIds":["evidence-provchange-meta-aasf-v2-0-moderate-threshold-later","evidence-provchange-meta-aasf-v2-0-moderate-threshold-earlier"],"attributes":{"changeBasis":"document_comparison","provision":{"kind":"threshold","key":"meta-moderate-threshold"}}},{"id":"sb-critical-incident-24-hours","statement":"SB 53 requires a frontier developer that discovers that a critical safety incident poses an imminent risk of death or serious physical injury to disclose that incident within 24 hours to an authority, including any law enforcement agency or public safety agency with jurisdiction, that is appropriate based on the nature of the incident and as required by law.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"What the enacted text states, quoted from the chaptered expression. It is not evidence that the statute applies to any named developer, or of compliance, enforcement or effect. The provision does not name a specific authority. Its relation to the 15-day report to OES in § 22757.13(c)(1), which is 'subject to paragraph (2)', is a question of legal interpretation and is not recorded here. No incident or disclosure is recorded.","evidenceIds":["evidence-sb-critical-incident-24-hours-1"],"attributes":{"provision":{"kind":"required_action","key":"sb53-imminent-risk-disclosure-24-hours"}}},{"id":"cs-openai-lp-announced-2019","statement":"OpenAI announced on 11 March 2019 that it was creating OpenAI LP as a hybrid of a for-profit and a nonprofit, which it called a “capped-profit” company. Investors and employees could receive a capped return, with the first round of investors capped at 100 times their investment, and returns beyond the cap would go to OpenAI Nonprofit.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"OpenAI’s own account of a structure it was creating. It is not evidence of the partnership’s formation date, its agreements or any payment of returns. The post says that “OpenAI” would from then on refer to OpenAI LP and the original entity to “OpenAI Nonprofit”, and that caps for later rounds would be negotiated per limited partner and expected to be lower. Control is recorded separately (the record of the March 2019 announcement on board control); the umbrella OpenAI identity is the subject because the post speaks for “OpenAI”.","evidenceIds":["evidence-cs-openai-lp-announced-2019-1","evidence-cs-openai-lp-announced-2019-2"],"attributes":{"control":{"instrument":"OpenAI LP announcement (11 March 2019)","stage":"announced"}}},{"id":"cm-wh-2023-07-inflection-ai","statement":"The White House named Inflection AI among the seven companies it convened on 21 July 2023, announcing that it had secured voluntary commitments from them to help move toward safe, secure and transparent development of AI technology.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The fact sheet links the July 2023 commitments text. Listing is the convener’s statement. It is not evidence that the organisation complied with, implemented or still maintains the commitments, and the legal effect of being listed is not recorded.","evidenceIds":["evidence-cm-wh-2023-07-inflection-ai-1"],"attributes":{"commitment":{"stage":"listed"}}},{"id":"cm-seoul-meta","statement":"Meta is listed among the sixteen organisations that, according to the GOV.UK text of 21 May 2024, the UK and Republic of Korea governments announced had agreed to the Frontier AI Safety Commitments.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Listing is the convener’s statement. It is not evidence that the organisation complied with, implemented or still maintains the commitments, and the legal effect of being listed is not recorded. Meta’s Frontier AI Framework v1.1 states that Meta signed the commitments in May 2024.","evidenceIds":["evidence-cm-seoul-meta-1"],"attributes":{"commitment":{"stage":"listed"}}},{"id":"ts-a4-hanson-1994-context","statement":"Hanson’s 1994 article discusses a sharp transition once uploads arrive.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"About uploads; no takeoff period.","evidenceIds":["evidence-ts-a4-hanson-1994-context"],"attributes":{"thought":{"roles":["context"],"register":"nonfiction","missing":"A sharp upload transition is discussed, but no period of hours to months from human level to far superhuman AI."}}},{"id":"ts-a5-burkhead-1996","statement":"Burkhead’s message of 1996-08-10 relays Good’s proposal of a computer programmed to design a still more intelligent computer.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A relay, not the poster’s own claim; the claimed 1979 printing was not checked.","evidenceIds":["evidence-ts-a5-burkhead-1996"],"attributes":{"thought":{"roles":["articulation"],"register":"relay","qualifier":"A relay of Good’s proposal."}}},{"id":"ts-a5-yudkowsky-2008-hard-takeoff","statement":"Yudkowsky’s post of 2008-12-02 compresses its argument into the slogan that it would be really odd if Moore’s Law stayed exponential with the improving computers doing the research.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Offered as a slogan, with the caveat that hardware is the least interesting factor.","evidenceIds":["evidence-ts-a5-yudkowsky-2008-hard-takeoff"],"attributes":{"thought":{"roles":["articulation"],"register":"nonfiction"}}},{"id":"lg-cn-jiang-daxin-stepfun-ceo-2026","statement":"36Kr, republishing a QbitAI report, identified Jiang Daxin on 27 January 2026 as StepFun’s CEO.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The record rests on reporting, not a company announcement; it does not date the start of the role. English rendering of the Chinese passage is a machine translation; no human translation review.","evidenceIds":["lg-ev-cn-jiang-daxin-stepfun-ceo-2026"],"attributes":{"role":"CEO"}},{"id":"lg-goo-corrado-dnn-project-founding-member","statement":"Greg Corrado's Google Research profile describes him as one of the founding members and the co-technical lead of Google's large-scale deep neural networks project.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The profile does not use the name Google Brain; identifying the project with Google Brain is not made here. It does not establish dates.","evidenceIds":["lg-ev-goo-corrado-dnn-project"],"attributes":{"role":"founding member and co-technical lead, Google's large scale deep neural networks project"}},{"id":"lg-gw-laskin-cofounded-reflection","statement":"Sequoia Capital’s announcement of 7 March 2025 names Misha Laskin as a co-founder of Reflection AI.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The record rests on an investor’s announcement of its partnership with the company; no Reflection AI page naming its founders was retrieved (its about, blog and newsroom pages do not name them). It does not establish the founding date or the completeness of the founder roster.","evidenceIds":["lg-ev-gw-laskin-cofounded-reflection"]},{"id":"ap-jeffrey-d-sachs-sachs-kotlikoff-2012-smart-machines","statement":"“Smart Machines and Long-Term Misery” (2012-12) credits Jeffrey D. Sachs as one of its 2 authors.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Credited co-authorship; it does not allocate parts of the text between authors or imply agreement with every passage.","evidenceIds":["evidence-ap-jeffrey-d-sachs-sachs-kotlikoff-2012-smart-machines"]},{"id":"ap-eliezer-yudkowsky-extropians-1997-09-26-yudkowsky-4420","statement":"The Extropians message “Re: Singularity: Generation gap” of 1997-09-26, archive message 4420, is sent under the name Eliezer S. Yudkowsky.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A sender name on a mailing-list message as archived; it does not establish identity beyond the credited name, and quoted text by others inside the message is not the sender’s.","evidenceIds":["evidence-ap-eliezer-yudkowsky-extropians-1997-09-26-yudkowsky-4420"]},{"id":"cpo-dc-fairwater-atlanta-unveiled","statement":"Microsoft unveiled on 12 November 2025 the next Fairwater site of Azure AI datacenters, in Atlanta, Georgia, connected to its Wisconsin Fairwater site.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The post states no capacity for the Atlanta site; its \"hundreds of thousands\" of GPUs describes the network design of Fairwater sites generally.","evidenceIds":["evidence-cpo-dc-fairwater-atlanta-unveiled-1","evidence-cpo-dc-fairwater-atlanta-unveiled-2"],"attributes":{"transactionStage":"announced"}},{"id":"ab-ev-o1-card-cb-medium-result","statement":"The o1 System Card states that its evaluations found o1 can help experts with the operational planning of reproducing a known biological threat, which it says meets OpenAI's medium risk threshold, and that the Safety Advisory Group classified the pre-mitigation model as medium risk for CBRN.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A risk classification stated by the developer for the model version it tested, under its own framework and thresholds. It is not an independent finding, is not generalised to other models or later versions, and says nothing about whether any safeguard works. The card adds that the exact performance numbers for the production model may vary. Only the summary and classification sentences are cited; the evaluation sections themselves hold technical content and are excluded.","evidenceIds":["evidence-ab-ev-o1-card-cb-medium-result-1","evidence-ab-ev-o1-card-cb-medium-result-2"],"attributes":{"evaluatedSystem":"o1, pre-mitigation and post-mitigation versions as described in the card","evaluation":{"what":"Chemical and biological threat creation capability, against OpenAI's own risk thresholds","how":"A suite of internal evaluations, with risk levels assigned by indicators and reviewed by the Safety Advisory Group","result":"Score for o1: Medium, the same as o1-preview; the Safety Advisory Group classified the pre-mitigation model as overall medium risk, including medium risk for CBRN, and rated post-mitigation risk the same","caveats":"The card states that the exact performance numbers for the model used in production may vary with final parameters, system prompt and other factors, and that the model tested was a near-final post-mitigation model."}}}],"sources":[{"id":"source-o1-card","title":"OpenAI o1 System Card","authors":["OpenAI"],"publisher":"OpenAI","publicationDate":"2024-12-05","version":"Page labelled Updated: December 5, 2024; includes multiple evaluation periods","versionDate":"2024-12-05","canonicalUrl":"https://openai.com/index/openai-o1-system-card/","evidenceUrl":"https://openai.com/index/openai-o1-system-card/","language":"en","archivedUrl":"https://web.archive.org/web/20260906142900/https://openai.com/index/openai-o1-system-card/","archivedAt":"2026-09-06","accessedAt":"2026-09-17","reuse":{"attribution":"OpenAI, OpenAI o1 System Card (2024-12-05).","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and minimal attributed excerpts only. Recheck terms before wider redistribution."}},{"id":"source-fsf-v3-update","title":"Strengthening our Frontier Safety Framework","authors":["Four Flynn","Helen King","Anca Dragan"],"publisher":"Google DeepMind","publicationDate":"2025-09-22","version":"Original v3.0 announcement; page updated April 17, 2026 with a separately labelled FSF 3.1 section","canonicalUrl":"https://deepmind.google/blog/strengthening-our-frontier-safety-framework/","evidenceUrl":"https://deepmind.google/blog/strengthening-our-frontier-safety-framework/","language":"en","archivedUrl":"https://web.archive.org/web/20251107005536/https://deepmind.google/blog/strengthening-our-frontier-safety-framework/","archivedAt":"2025-11-07","accessedAt":"2026-09-18","reuse":{"attribution":"Google DeepMind, Strengthening our Frontier Safety Framework.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-rsp-3-0","title":"Responsible Scaling Policy — v3.0","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":"2026-02-24","version":"Responsible Scaling Policy v3.0; effective date 2026-02-24.","versionDate":"2026-02-24","canonicalUrl":"https://anthropic.com/responsible-scaling-policy/rsp-v3-0","evidenceUrl":"https://anthropic.com/responsible-scaling-policy/rsp-v3-0","language":"en","archivedUrl":"https://web.archive.org/web/20260224203046/https://www.anthropic.com/responsible-scaling-policy/rsp-v3-0","archivedAt":"2026-02-24","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v3.0.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-rsp-3-1","title":"Responsible Scaling Policy — v3.1","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Responsible Scaling Policy v3.1; effective date 2026-04-02.","versionDate":"2026-04-02","canonicalUrl":"https://www-cdn.anthropic.com/files/4zrzovbb/website/bf04581e4f329735fd90634f6a1962c13c0bd351.pdf","evidenceUrl":"https://www-cdn.anthropic.com/files/4zrzovbb/website/bf04581e4f329735fd90634f6a1962c13c0bd351.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260407193305/https://www-cdn.anthropic.com/files/4zrzovbb/website/bf04581e4f329735fd90634f6a1962c13c0bd351.pdf","archivedAt":"2026-04-07","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v3.1.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-rsp-3-2","title":"Responsible Scaling Policy — v3.2","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Responsible Scaling Policy v3.2; effective date 2026-04-29.","versionDate":"2026-04-29","canonicalUrl":"https://cdn.sanity.io/files/4zrzovbb/website/28c6241900d90410628a8a2003a5572faae4365a.pdf","evidenceUrl":"https://cdn.sanity.io/files/4zrzovbb/website/28c6241900d90410628a8a2003a5572faae4365a.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260505141510/https://cdn.sanity.io/files/4zrzovbb/website/28c6241900d90410628a8a2003a5572faae4365a.pdf","archivedAt":"2026-05-05","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v3.2.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-fsf-1-0","title":"Frontier Safety Framework — v1.0","authors":["Google DeepMind"],"publisher":"Google DeepMind","publicationDate":"2024-05-17","version":"Frontier Safety Framework v1.0; publication date 2024-05-17.","versionDate":"2024-05-17","canonicalUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/introducing-the-frontier-safety-framework/fsf-technical-report.pdf","evidenceUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/introducing-the-frontier-safety-framework/fsf-technical-report.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20240517140902/https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/introducing-the-frontier-safety-framework/fsf-technical-report.pdf","archivedAt":"2024-05-17","accessedAt":"2026-09-18","reuse":{"attribution":"Google DeepMind, Frontier Safety Framework — v1.0.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-eval-i-o1-preview","title":"Details about METR’s preliminary evaluation of OpenAI o1-preview","authors":["METR"],"publisher":"METR","publicationDate":"2024-09-12","version":"Dated preliminary report, current migrated web expression inspected 2026-09-27.","canonicalUrl":"https://metr.org/evaluations/openai-o1-preview-report/","evidenceUrl":"https://metr.org/evaluations/openai-o1-preview-report/","language":"en","archivedUrl":"https://web.archive.org/web/20260914201621/https://metr.org/evaluations/openai-o1-preview-report/","archivedAt":"2026-09-14","accessedAt":"2026-09-27","reuse":{"attribution":"METR, Details about METR’s preliminary evaluation of OpenAI o1-preview.","licence":"No open redistribution licence asserted","note":"Attributed paraphrases and bibliographic metadata only; original documents linked, not redistributed."}},{"id":"source-atlas-openai-lp","title":"OpenAI LP","authors":["Greg Brockman","Ilya Sutskever","OpenAI"],"publisher":"OpenAI","publicationDate":"2019-03-11","version":"Dated publisher page inspected 2026-09-28 and again 2026-10-04 at its current address; the quotes used from 2026-10-04 also match the Internet Archive capture of 2019-03-11 (16:15 UTC) at the original address, which uses straight quotation marks and an additional standfirst.","canonicalUrl":"https://openai.com/index/openai-lp/","evidenceUrl":"https://openai.com/index/openai-lp/","language":"en","archivedUrl":"https://web.archive.org/web/20190311161512/https://openai.com/blog/openai-lp/","archivedAt":"2019-03-11","accessedAt":"2026-10-04","reuse":{"attribution":"Greg Brockman; Ilya Sutskever; OpenAI. OpenAI LP.","licence":"No redistribution licence asserted.","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-xm-llama-paper","title":"LLaMA: Open and Efficient Foundation Language Models","authors":["Hugo Touvron","Thibaut Lavril","Gautier Izacard","Xavier Martinet","Marie-Anne Lachaux","Timothée Lacroix","Baptiste Rozière","Naman Goyal","Eric Hambro","Faisal Azhar","Aurelien Rodriguez","Armand Joulin","Edouard Grave","Guillaume Lample"],"publisher":"arXiv","publicationDate":"2023-02-27","version":"arXiv:2302.13971v1; first submitted version.","canonicalUrl":"https://arxiv.org/html/2302.13971v1","evidenceUrl":"https://arxiv.org/html/2302.13971v1","language":"en","accessedAt":"2026-09-29","reuse":{"attribution":"LLaMA: Open and Efficient Foundation Language Models — arXiv","licence":"No redistribution licence asserted.","note":"Source-linked metadata and brief evidence paraphrases; original text remains with its publisher."}},{"id":"source-meta-faf-v1-1","title":"Frontier AI Framework — Version 1.1","authors":["Meta"],"publisher":"Meta","publicationDate":"2025-02-03","version":"Frontier AI Framework, Version 1.1, as captured by the Internet Archive on 2025-02-13, the earliest complete capture located. The original address has returned HTTP 500 since April 2026; a revised expression under the same label was captured from 2025-03-28 and is recorded separately.","versionDate":"2025-02-03","canonicalUrl":"https://ai.meta.com/static-resource/meta-frontier-ai-framework/","evidenceUrl":"https://web.archive.org/web/20250213105620/https://ai.meta.com/static-resource/meta-frontier-ai-framework/","language":"en","archivedUrl":"https://web.archive.org/web/20250213105620/https://ai.meta.com/static-resource/meta-frontier-ai-framework/","archivedAt":"2025-02-13","accessedAt":"2026-10-03","reuse":{"attribution":"Meta, Frontier AI Framework, Version 1.1 (2025).","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-meta-aasf-v2","title":"Advanced AI Scaling Framework — Version 2","authors":["Meta"],"publisher":"Meta","publicationDate":"2026-04-08","version":"Advanced AI Scaling Framework, Version 2; its change log dates the version April 7, 2026. The file served at the recorded address is byte-identical to the Internet Archive capture of 2026-04-08.","versionDate":"2026-04-07","canonicalUrl":"https://ai.meta.com/static-resource/Meta_Advanced-AI-Scaling-Framework-v2","evidenceUrl":"https://ai.meta.com/static-resource/Meta_Advanced-AI-Scaling-Framework-v2","language":"en","archivedUrl":"https://web.archive.org/web/20260408163104/https://ai.meta.com/static-resource/Meta_Advanced-AI-Scaling-Framework-v2","archivedAt":"2026-04-08","accessedAt":"2026-10-03","reuse":{"attribution":"Meta, Advanced AI Scaling Framework, Version 2 (2026).","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-ca-sb53-chaptered","title":"SB-53 Artificial intelligence models: large developers. (2025-2026) — bill text, chaptered version (Chapter 138, Statutes of 2025)","authors":["California Legislature"],"publisher":"California State Legislature","publicationDate":"2025-09-29","version":"Chaptered version dated 09/29/25, inspected live 2026-10-04; quote also matched in the 2025-09-30 capture. Publisher as printed on the inspected expression: California Legislature (California Legislative Information, Office of Legislative Counsel).","canonicalUrl":"https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB53","evidenceUrl":"https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB53","language":"en","archivedUrl":"https://web.archive.org/web/20250930154105/https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB53","archivedAt":"2025-09-30","accessedAt":"2026-10-04","reuse":{"attribution":"California Legislature, SB-53 Artificial intelligence models: large developers. (2025-2026) — bill text, chaptered version (Chapter 138, Statutes of 2025).","licence":"Official government text; no redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-wh-factsheet-2023-07","title":"FACT SHEET: Biden-Harris Administration Secures Voluntary Commitments from Leading Artificial Intelligence Companies to Manage the Risks Posed by AI","authors":["The White House"],"publisher":"The White House","publicationDate":"2023-07-21","version":"Fact sheet dated July 21, 2023, inspected 2026-10-04 at its archived-site address; quotes also matched in the Internet Archive capture of 2023-07-21 (10:53 UTC) at the original address. The Biden–Harris White House pages are now served from the archived site bidenwhitehouse.archives.gov.","canonicalUrl":"https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai/","evidenceUrl":"https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai/","language":"en","archivedUrl":"https://web.archive.org/web/20230721105339/https://www.whitehouse.gov/briefing-room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai/","archivedAt":"2023-07-21","accessedAt":"2026-10-04","reuse":{"attribution":"The White House, fact sheet of 21 July 2023 on voluntary AI commitments.","licence":"US government work; no redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-fasc-2024","title":"Frontier AI Safety Commitments, AI Seoul Summit 2024","authors":["Department for Science, Innovation and Technology"],"publisher":"Department for Science, Innovation and Technology","publicationDate":"2024-05-21","version":"GOV.UK HTML publication, first published 21 May 2024 and updated 7 February 2025 to add organisations to the list; inspected live 2026-10-04. The Internet Archive capture of 2024-05-21 (11:05 UTC) has the sixteen-name list only, and the commitment text quoted is identical in both. Publisher as printed on the inspected expression: Department for Science, Innovation and Technology (GOV.UK).","versionDate":"2025-02-07","canonicalUrl":"https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024","evidenceUrl":"https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024","language":"en","archivedUrl":"https://web.archive.org/web/20240521110530/https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024","archivedAt":"2024-05-21","accessedAt":"2026-10-04","reuse":{"attribution":"Department for Science, Innovation and Technology, Frontier AI Safety Commitments, AI Seoul Summit 2024 (GOV.UK).","licence":"Open Government Licence v3.0 (GOV.UK content); quoted briefly under the editorial policy","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-extropy-13-scan","title":"Extropy, no. 13 (scan), \"If Uploads Come First\"","authors":["Robin Hanson"],"publisher":"Extropy","publicationDate":"1994","version":"Photographic scan of the issue; printed pp. 10–15 (PDF pages 8–13) read by the pre-1996 candidate search and a machine check, which confirmed the copyright header and footers. Copy hosted at diyhpl.us.","canonicalUrl":"https://diyhpl.us/~bryan/irc/extropians/extropy-magazine/Extropy-13.pdf","evidenceUrl":"https://diyhpl.us/~bryan/irc/extropians/extropy-magazine/Extropy-13.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250428112724/https://diyhpl.us/~bryan/irc/extropians/extropy-magazine/Extropy-13.pdf","archivedAt":"2025-04-28","accessedAt":"2026-10-06","reuse":{"attribution":"Robin Hanson, \"If Uploads Come First\", Extropy 13 (1994).","licence":"No open redistribution licence asserted","note":"Bibliographic metadata and labelled paraphrases only: the inspected scan is a third-party copy whose permission to host is not established. Phrases in quotation marks inside a paraphrase identify the source's wording."}},{"id":"source-extropians-1996-08-10-burkhead","title":"intelligence increase (Extropians mailing list)","authors":["Lyle Burkhead"],"publisher":"Extropians mailing list","publicationDate":"1996-08-10","version":"Archive HTML as served, retrieved 2026-10-06 with the hash recorded by the 1996–2022 candidate search; quoted lines excluded when reading the author’s text. Archive at extropians.weidai.com.","canonicalUrl":"http://extropians.weidai.com/extropians.96/0220.html","evidenceUrl":"http://extropians.weidai.com/extropians.96/0220.html","language":"en","archivedUrl":"https://web.archive.org/web/20220523234349/https://extropians.weidai.com/extropians.96/0220.html","archivedAt":"2022-05-23","accessedAt":"2026-10-06","reuse":{"attribution":"Lyle Burkhead, \"intelligence increase\", Extropians mailing list, 1996-08-10.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-yudkowsky-2008-hard-takeoff","title":"Hard Takeoff","authors":["Eliezer Yudkowsky"],"publisher":"Overcoming Bias","publicationDate":"2008-12-02","version":"Text checked in the post’s first stored version through the LessWrong GraphQL API on 2026-10-06; content version 1.0.0. Now hosted on LessWrong.","canonicalUrl":"https://www.lesswrong.com/posts/tjH8XPxAnr6JRbh7k/hard-takeoff","evidenceUrl":"https://www.lesswrong.com/posts/tjH8XPxAnr6JRbh7k/hard-takeoff","language":"en","archivedUrl":"https://web.archive.org/web/20181213042057/https://www.lesswrong.com/posts/tjH8XPxAnr6JRbh7k/hard-takeoff","archivedAt":"2018-12-13","accessedAt":"2026-10-06","reuse":{"attribution":"Eliezer Yudkowsky, \"Hard Takeoff\", Overcoming Bias (now hosted on LessWrong), 2008-12-02.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-extropians-1997-09-26-yudkowsky-4420","title":"Re: Singularity: Generation gap (Extropians mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"Extropians mailing list","publicationDate":"1997-09-26","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 4553a5e84750edd9528beb9e2c297772caee01b17f1653ee34bdae4f8ca62354); quoted lines excluded when reading the author’s text. The served archive shows later URL rewriting and does not authenticate historical bytes. Archive at extropians.weidai.com.","canonicalUrl":"http://extropians.weidai.com/extropians.3Q97/4420.html","evidenceUrl":"http://extropians.weidai.com/extropians.3Q97/4420.html","language":"en","archivedUrl":"https://web.archive.org/web/20261005144258/https://extropians.weidai.com/extropians.3Q97/4420.html","archivedAt":"2026-10-05","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Re: Singularity: Generation gap”, Extropians mailing list, 1997-09-26.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-sachs-kotlikoff-2012-smart-machines","title":"Smart Machines and Long-Term Misery","authors":["Jeffrey D. Sachs","Laurence J. Kotlikoff"],"publisher":"NBER Working Paper 18629","publicationDate":"2012-12","version":"Full text. Current copy retrieved directly on 2026-10-09; no archive capture (Wayback unreachable), so earlier bytes are not authenticated. SHA-256 aa4958092702ecfad60c57fcdff3a76505353fccab0bcb60bda2fb1b14c0a65b.","canonicalUrl":"https://www.nber.org/system/files/working_papers/w18629/w18629.pdf","evidenceUrl":"https://www.nber.org/system/files/working_papers/w18629/w18629.pdf","language":"en","accessedAt":"2026-10-09","reuse":{"attribution":"Jeffrey D. Sachs, Laurence J. Kotlikoff, “Smart Machines and Long-Term Misery”, NBER Working Paper 18629, 2012-12.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"lg-src-cn-qbitai-stepfun-2026","title":"阶跃星辰不再低调：巨额融资，印奇加入，“1+3”核心决策层浮出水面","authors":["量子位 (QbitAI)"],"publisher":"36氪 (36Kr), republishing 量子位","publicationDate":"2026-01-27","version":"Dated page, retrieved on 2026-10-15 (SHA-256 8081f6283cd0326b9aeb5becc09aa65187d5df211d0d94ab82444c68c7631473) directly from the publisher.","canonicalUrl":"https://www.36kr.com/p/3657567182938757","evidenceUrl":"https://www.36kr.com/p/3657567182938757","language":"zh","archivedUrl":"https://web.archive.org/web/20261029003800/https://www.36kr.com/p/3657567182938757","archivedAt":"2026-10-29","accessedAt":"2026-10-15","reuse":{"attribution":"量子位 (QbitAI), “阶跃星辰不再低调：巨额融资，印奇加入，“1+3”核心决策层浮出水面”, 2026-01-27.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"lg-src-goo-research-corrado","title":"Greg Corrado (Google Research profile, undated)","authors":["Google Research"],"publisher":"Google Research","publicationDate":null,"version":"Undated page, retrieved on 2026-10-15 (SHA-256 6a31afc118f931926bc64a7877196952f8d7b6fa204a9c718b48e06a2375ac5a) directly from the publisher. Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://research.google/people/gregcorrado/","evidenceUrl":"https://research.google/people/gregcorrado/","language":"en","archivedUrl":"https://web.archive.org/web/20260925005957/https://research.google/people/gregcorrado/","archivedAt":"2026-09-25","accessedAt":"2026-10-15","reuse":{"attribution":"Google Research, “Greg Corrado (Google Research profile, undated)”.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"lg-src-gw-sequoia-reflection-2025","title":"Partnering with Reflection: Toward Superintelligence, with Autonomous Coding","authors":["Stephanie Zhan","Charlie Curnin"],"publisher":"Sequoia Capital","publicationDate":"2025-03-07","version":"Dated page, retrieved on 2026-10-15 (SHA-256 5089a82bf34f432446896e540cfaee45f5bbbb9b8ff708724fe3f33578406dd0) directly from the publisher.","canonicalUrl":"https://sequoiacap.com/article/partnering-with-reflection-toward-superintelligence-with-autonomous-coding","evidenceUrl":"https://sequoiacap.com/article/partnering-with-reflection-toward-superintelligence-with-autonomous-coding","language":"en","archivedUrl":"https://web.archive.org/web/20261010210147/https://sequoiacap.com/article/partnering-with-reflection-toward-superintelligence-with-autonomous-coding","archivedAt":"2026-10-10","accessedAt":"2026-10-15","reuse":{"attribution":"Stephanie Zhan, Charlie Curnin, “Partnering with Reflection: Toward Superintelligence, with Autonomous Coding”, 2025-03-07.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-cpo-dc-msft-ai-superfactory-2025","title":"Infinite scale: The architecture behind the Azure AI superfactory","authors":["Microsoft (Official Microsoft Blog; Scott Guthrie, Executive Vice President, Cloud + AI)"],"publisher":"Microsoft (Official Microsoft Blog; Scott Guthrie, Executive Vice President, Cloud + AI)","publicationDate":"2025-11-12","version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 29fd762caa98a590efe51391fc4cc752ed2f0006cb68cd2f036ece897eb45026). Blog post as served on 2026-10-28; byline \"Nov 12, 2025 | Scott Guthrie\". The page carries an editor's note of a later update (not dated in the retrieved text). Retrieved again at admission on 2026-10-30 (SHA-256 29fd762caa98a590efe51391fc4cc752ed2f0006cb68cd2f036ece897eb45026, the same bytes). Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://blogs.microsoft.com/blog/2025/11/12/infinite-scale-the-architecture-behind-the-azure-ai-superfactory/","evidenceUrl":"https://blogs.microsoft.com/blog/2025/11/12/infinite-scale-the-architecture-behind-the-azure-ai-superfactory/","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"Microsoft (Official Microsoft Blog; Scott Guthrie, Executive Vice President, Cloud + AI), “Infinite scale: The architecture behind the Azure AI superfactory”, 2025-11-12.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}}],"evidence":[{"id":"evidence-fsf-v3-harmful-manipulation","sourceRole":"primary","support":"supports","locator":"Section \"Addressing the risks of harmful manipulation\" (original v3.0 section), first sentence","passage":"we’re introducing a Critical Capability Level (CCL)* focused on harmful manipulation — specifically, AI models with powerful manipulative capabilities that could be misused to systematically and substantially change beliefs and behaviors in identified high stakes contexts","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-03 against the live page and the Internet Archive capture of 2025-11-07, which predates the appended FSF 3.1 section; it replaces an earlier attributed paraphrase. Human editorial review is pending."},{"id":"evidence-eval-i-o1-method","sourceRole":"primary","support":"supports","locator":"Details of Evaluation Procedure > Descriptions of task suites, numbered item 1, general-autonomy test suite.","passage":"METR says its general-autonomy test suite is identical to the suite in its earlier GPT-4o report.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-27","note":"Original publisher passage machine-inspected. This is not human review or independent reproduction of the reported activity."},{"id":"evidence-xm-touvron-llama","sourceRole":"primary","support":"supports","locator":"Version-one title-page author list.","passage":"Hugo Touvron appears in the author list.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-29","note":"Passage checked by machine against the identified source expression. No human editorial acceptance or independent verification of implementation is implied."},{"id":"evidence-prov-rsp-v3-2-novel-cb-threshold","sourceRole":"primary","support":"supports","locator":"PDF p. 7 (printed p. 7), §1 Our Recommendations for Industry-Wide Safety, table row 'Novel chemical/biological weapons production', left column ('Capability or usage threshold')","passage":"AI systems with the ability to significantly help threat actors (for example, moderately resourced expert-backed teams) create/obtain and deploy chemical and/or biological weapons with potential for catastrophic damages far beyond those of past catastrophes such as COVID-19.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-03 against the inspected PDF, byte-identical to the Internet Archive capture of 2026-05-05. Extraction note: Quote omits the row label. Wording identical in v3.0 and v3.1 (PDF p. 7). Human editorial review is pending."},{"id":"evidence-prov-fsf-2024-alert-threshold","sourceRole":"primary","support":"supports","locator":"PDF p. 2 (printed p. 2), Framework, component “2 - Evaluating frontier models”, first paragraph, second sentence","passage":"we will define a set of evaluations called “early warning evaluations,” with a specific “pass” condition that flags when a CCL may be reached before the evaluations are run again.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-03 against the inspected PDF, byte-identical to the Internet Archive capture of 2024-05-17. Extraction note: Quote starts mid-sentence after “To do so,”; meaning unchanged. Human editorial review is pending."},{"id":"evidence-absent-rsp-v3-1-no-train-deploy-without-safeguards","sourceRole":"primary","support":"supports","locator":"RSP v3.1 (effective April 2, 2026), whole text layer, PDF pp. 1–20: Introduction (pp. 3–4); §1 with the threshold table (pp. 4–10); §2–§3 Risk Reports (pp. 11–15); §4 Governance (p. 16); Appendix A (pp. 16–17); Appendix B (p. 17); Changelog (pp. 17–20).","passage":"Within the inspected scope, no passage stating the provision “Commitment not to train or deploy without adequate safeguards” was found.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-03","note":"Absence recorded by an automated research assistant on 2026-10-03. Method: Same reading and comparison as for v3.0, with the same case-insensitive search terms. No passage stating a commitment not to train or deploy models unless adequate safeguards are implemented, or a commitment to pause training or deployment until such safeguards are in place, was found within the operative text. Hits examined and not counted: the same Introduction, §1 and Appendix A passages as in v3.0 (Appendix A table p. 17); the reproduced RSP v2.0 changelog entry (p. 17); and two sentences new in the Appendix A introduction (p. 16) stating that the competitor commitments do not preclude cautionary action such as refraining from training or deploying models in other circumstances, and that Anthropic would strongly consider pausing development and/or deployment to improve its models' safety profiles even in cases not covered. These state possible or considered actions, not a commitment conditioned on safeguards; the v3.1 changelog entry, item (3) (p. 20), describes them as a statement that Anthropic may consider pausing. Human editorial review is pending."},{"id":"evidence-prov-rsp-v3-0-competitor-parity","sourceRole":"primary","support":"supports","locator":"PDF p. 16 (printed p. 16), Appendix A: Commitments Related to Competitors, table row 'Competitors have strong safety measures', Commitment column, first sentence","passage":"For our highly capable frontier models, we will meet or exceed the overall risk reduction posture of these competitors, as far as we can tell based on our best efforts to assess that posture.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-03 against the PDF served through the recorded URL’s redirect, byte-identical to the Internet Archive capture of 2026-02-24. Extraction note: Scenario column: strong evidence that all competitors who have developed or will soon develop a highly capable frontier model can make strong arguments that catastrophic risk is contained. The cell's second sentence, which commits to delay AI development and deployment as needed until Anthropic is able to meet that posture, is quoted as a separate passage (2026-10-25); the whole cell is 53 words. Same wording in v3.1–v3.4 (PDF p. 17); not recorded there (bounded). Human editorial review is pending."},{"id":"evidence-prov-rsp-v3-0-competitor-parity-delay","sourceRole":"primary","support":"supports","locator":"PDF p. 16 (printed p. 16), Appendix A: Commitments Related to Competitors, table row 'Competitors have strong safety measures', Commitment column, second sentence","passage":"Until we are able to do so, we will delay AI development and deployment as needed to achieve this.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-25","note":"Verbatim excerpt checked on 2026-10-25 against the publisher’s pinned v3.0 PDF (www-cdn.anthropic.com/e670587677525f28df69b59e5fb4c22cc5461a17.pdf, SHA-256 a71bfa08a7fa4b750b2b408b4b9c8681ad50c6b53271c314f591c59c84c38064), which the cross-family check of 2026-10-25 found byte-identical to the Internet Archive capture of 2026-02-24. A stated commitment, not evidence of practice or compliance. Human editorial review is pending."},{"id":"evidence-provchange-meta-aasf-v2-0-moderate-threshold-later","sourceRole":"primary","support":"supports","locator":"PDF p. 15 (printed p. 15), §3.3 Risk Thresholds, Table 1, 'Moderate or lower' row","passage":"Advanced AI Scaling Framework v2.0 states, in the 'Moderate or lower' row of Table 1, that the Frontier AI shows relevant capabilities but could not substantially contribute to any threat scenario associated with a catastrophic outcome across plausible deployment and development scenarios.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-03","note":"Paraphrase checked by an automated research assistant on 2026-10-03 against the cited edition; the edition’s own wording is quoted in its provision record. Human editorial review is pending."},{"id":"evidence-provchange-meta-aasf-v2-0-moderate-threshold-earlier","sourceRole":"primary","support":"supports","locator":"PDF p. 13 (printed p. 13), §3.3 Risk thresholds, Table 1, Moderate row","passage":"Frontier AI Framework v1.1 states, in the Moderate row of Table 1, that the model would not provide significant uplift towards execution of a threat scenario.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-03","note":"Paraphrase checked by an automated research assistant on 2026-10-03 against the cited edition; the edition’s own wording is quoted in its provision record. Human editorial review is pending."},{"id":"evidence-sb-critical-incident-24-hours-1","sourceRole":"primary","support":"supports","locator":"SEC. 2, Bus. & Prof. Code § 22757.13(c)(2)","passage":"the frontier developer shall disclose that incident within 24 hours to an authority, including any law enforcement agency or public safety agency with jurisdiction, that is appropriate based on the nature of that incident and as required by law","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-04","note":"Verbatim excerpt proposed by a research agent and checked by two independent fresh-context verifier agents and decided by a third adjudicating agent on 2026-10-04; re-matched mechanically by the integrator against a direct fetch of the cited expression. 39 words. Extended in adjudication to include the statute's description of the authority. The statement uses the same paragraph's wording for the trigger: 'poses an imminent risk of death or serious physical injury'. Human editorial review is pending."},{"id":"evidence-cs-openai-lp-announced-2019-1","sourceRole":"primary","support":"supports","locator":"Introduction, paragraph beginning “We want to increase our ability to raise capital”","passage":"create OpenAI LP as a hybrid of a for-profit and nonprofit","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-04","note":"Verbatim quote, re-checked on 2026-10-04 by the integrating session against its own direct fetch of the cited expression (mechanical match after whitespace normalisation). Also matched in the Internet Archive capture of 2019-03-11, where the sentence continues with straight quotation marks around “capped-profit”. The integrating session wrote or selected this passage from its own reading of the source; it does not repeat a passage from the week 4b research pass. Human editorial review is pending."},{"id":"evidence-cs-openai-lp-announced-2019-2","sourceRole":"primary","support":"supports","locator":"Section “The mission comes first”, paragraph beginning “As mentioned above”","passage":"Returns for our first round of investors are capped at 100x their investment","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-04","note":"Verbatim quote, re-checked on 2026-10-04 by the integrating session against its own direct fetch of the cited expression (mechanical match after whitespace normalisation). Also matched in the Internet Archive capture of 2019-03-11. The passage was first proposed and machine-verified in the week 4b research pass. Human editorial review is pending."},{"id":"evidence-cm-wh-2023-07-inflection-ai-1","sourceRole":"primary","support":"supports","locator":"First body paragraph after the two subheadings","passage":"Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-04","note":"Verbatim quote, re-checked on 2026-10-04 by the integrating session against its own direct fetch of the cited expression (mechanical match after whitespace normalisation). Also matched in the Internet Archive capture of 2023-07-21. The passage was first proposed and machine-verified in the week 4b research pass. Human editorial review is pending."},{"id":"evidence-cm-seoul-meta-1","sourceRole":"primary","support":"supports","locator":"Opening sentence and first list of organisations (sixteen names)","passage":"The text opens by stating that the UK and Republic of Korea governments announced that the listed organisations have agreed to the commitments, and the list includes “Meta”.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-04","note":"Editorial paraphrase, re-checked on 2026-10-04 by the integrating session against its own direct fetch of the cited expression. The list is identical in the Internet Archive capture of 2024-05-21. The integrating session wrote or selected this passage from its own reading of the source; it does not repeat a passage from the week 4b research pass. Human editorial review is pending."},{"id":"evidence-ts-a4-hanson-1994-context","sourceRole":"primary","support":"supports","locator":"Printed pp. 12–14 (PDF pages 10–12)","passage":"The article discusses how sudden the transition could be once uploading becomes possible, without a period from human level to far superhuman AI.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-06","note":"Paraphrase from the pre-1996 candidate search."},{"id":"evidence-ts-a5-burkhead-1996","sourceRole":"primary","support":"supports","locator":"Message body, first paragraph of the reposted letter","passage":"Good proposes [...] a computer more intelligent than a man [...] programmed to design a still more intelligent computer","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-06","note":"Checked against the archive HTML on 2026-10-06."},{"id":"evidence-ts-a5-yudkowsky-2008-hard-takeoff","sourceRole":"primary","support":"supports","locator":"Paragraph after the discussion of the soft-takeoff keyhole","passage":"Moore's Law is exponential now; it would be really odd if it stayed exponential with the improving computers doing the research.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-28","note":"Checked against the first stored version on 2026-10-06; quoted in the 1996–2022 candidate search’s note but not as its passage. On 2026-10-30, by the maintainer’s decision of that day, the quotation was corrected to the archive capture of 2018-12-13 (quotation marks and apostrophes as in the capture), which a cross-family archive check read on 2026-10-28; the proposition is unchanged."},{"id":"lg-ev-cn-jiang-daxin-stepfun-ceo-2026","sourceRole":"secondary","support":"supports","locator":"Paragraph listing the “1+3” group","passage":"分别是阶跃星辰CEO姜大昕，首席科学家张祥雨，CTO朱亦博","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (1 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read directly from the publisher. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-16/17) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"lg-ev-goo-corrado-dnn-project","sourceRole":"primary","support":"supports","locator":"Profile biography, final sentence","passage":"one of the founding members and the co-technical lead of Google's large scale deep neural networks project","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (17 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read directly from the publisher. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. No human editorial review.","retrievalRoute":"direct"},{"id":"lg-ev-gw-laskin-cofounded-reflection","sourceRole":"secondary","support":"supports","locator":"Photo caption below the byline","passage":"Reflection co-founders Misha Laskin and Ioannis Antonoglou.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (7 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read directly from the publisher. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-16/17) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ap-jeffrey-d-sachs-sachs-kotlikoff-2012-smart-machines","sourceRole":"primary","support":"supports","locator":"PDF text layer, first two pages: author name on the title page or first page","passage":"Jeffrey D. Sachs","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-27","note":"Credited name checked by an automated research assistant on 2026-10-27 in a direct retrieval and checked by a reviewer from a different model family on 2026-10-27. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"evidence-ap-eliezer-yudkowsky-extropians-1997-09-26-yudkowsky-4420","sourceRole":"primary","support":"supports","locator":"Message header, sender name (address not reproduced)","passage":"Eliezer S. Yudkowsky","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-27","note":"Credited name checked by an automated research assistant on 2026-10-27 in a direct retrieval and checked by a reviewer from a different model family on 2026-10-27. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"evidence-cpo-dc-fairwater-atlanta-unveiled-1","sourceRole":"primary","support":"supports","locator":"paragraph 1","passage":"Today, we are unveiling the next Fairwater site of Azure AI datacenters in Atlanta, Georgia.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, compute, power and orbit (research/compute-power-orbit). The quotation (15 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-cpo-dc-fairwater-atlanta-unveiled-2","sourceRole":"primary","support":"supports","locator":"paragraph 1","passage":"the Atlanta datacenter is connected to Microsoft's first Fairwater site in Wisconsin.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, compute, power and orbit (research/compute-power-orbit). The paraphrase is Selekoa’s, written from the text retrieved for the candidate pack on 2026-10-28; it is not a quotation and was not machine-matched. The cross-family check found it faithful. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-o1-card-cb-medium-result-1","sourceRole":"primary","support":"supports","locator":"PDF p. 19, §5.5 Chemical and Biological Threat Creation, summary line","passage":"Our evaluations found that o1 can help experts with the operational planning of reproducing a known biological threat, which meets our medium risk threshold.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (24 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-30, item 23) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-o1-card-cb-medium-result-2","sourceRole":"primary","support":"supports","locator":"PDF p. 16, §5 Preparedness Framework Evaluations, paragraph on the Safety Advisory Group","passage":"the Safety Advisory Group … classified the o1 pre-mitigation model as overall medium risk, including medium risk for persuasion and CBRN, and low risk for model autonomy and cybersecurity.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (28 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. The ellipsis stands for a reference marker in the PDF text layer. A cross-family machine check (2026-10-30, item 23) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"}],"at":{"entities":[0,4,19,22,26,39,45,46,47,50,51,115,116,157,161,165,199,200,205,208,235,250,252,282,285,317,339,340,394,536,677,701,746,773,803,918,1149],"claims":[96,267,342,420,429,486,539,600,613,627,676,707,850,858,928,1840,1965,1990,2320,2357,3140,3375],"sources":[13,32,39,40,41,44,116,121,152,178,180,191,237,239,284,286,322,361,487,584,646,684,971],"evidence":[113,288,365,452,461,517,605,606,679,680,709,747,748,853,886,1064,1072,1150,2367,2538,2570,2967,3004,4159,4160,4635,4636]},"version":"1lg47t7"}