{"entities":[{"id":"openai","name":"OpenAI"},{"id":"deepmind","name":"DeepMind"},{"id":"rsp-2023","name":"Responsible Scaling Policy (2023)"},{"id":"fsf-2024","name":"Frontier Safety Framework (2024)"},{"id":"google","name":"Google"},{"id":"rsp-v3-0","name":"Anthropic Responsible Scaling Policy — v3.0"},{"id":"fsf-v3-0","name":"Google DeepMind Frontier Safety Framework — v3.0"},{"id":"fsf-v3-1","name":"Google DeepMind Frontier Safety Framework — v3.1"},{"id":"daniel-gross","name":"Daniel Gross"},{"id":"ssi","name":"Safe Superintelligence Inc."},{"id":"weak-strong-paper","name":"Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision"},{"id":"paul-christiano","name":"Paul Christiano"},{"id":"ms-fgf-2025","name":"Microsoft Frontier Governance Framework — Version 1 (February 2025)"},{"id":"openai-group-pbc-certificate-2025","name":"Certificate of Incorporation of OpenAI Group PBC (filed 2025-10-28)"},{"id":"dsit","name":"Department for Science, Innovation and Technology"},{"id":"mistral-ai","name":"Mistral AI"},{"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-1996-09-03-hanson","name":"Re: Thinking about the future..."},{"id":"fast-or-hard-takeoff","name":"Fast or hard takeoff"},{"id":"david-luan","name":"David Luan"},{"id":"fidji-simo","name":"Fidji Simo"},{"id":"robin-hanson","name":"Robin Hanson"},{"id":"aiia-ai-safety-commitments-2024","name":"Artificial Intelligence Safety Commitments (人工智能安全承诺), December 2024"},{"id":"ant-group","name":"Ant Group / 蚂蚁集团"},{"id":"us-ai-action-plan-2025","name":"Winning the Race: America’s AI Action Plan"},{"id":"claude-3-7-sonnet-system-card","name":"Claude 3.7 Sonnet System Card"}],"claims":[{"id":"google-acquired-deepmind-2014","statement":"Google acquired DeepMind in early 2014.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"No purchase price, legal-entity identifier, or 2023 combination is implied.","evidenceIds":["evidence-google-acquired-deepmind-2014"]},{"id":"daniel-gross-ssi-cofounder-2024","statement":"Reuters identifies Daniel Gross as a co-founder of Safe Superintelligence in its June 2024 launch report.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Contemporaneous secondary reporting; original founding page changed and archive retrieval failed. No OpenAI-to-SSI corporate descent inferred.","evidenceIds":["evidence-daniel-gross-ssi-cofounder-2024"]},{"id":"lineage-ii-openai-published-weak","statement":"OpenAI’s Weak-to-strong generalization announcement presents the linked paper as the Superalignment team’s first paper.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Publisher’s attribution of programme output. Mutable expression displaying 2023-12-14, not authenticated wording from that day; not independent verification of results or derivation from Good/MIRI.","evidenceIds":["evidence-lineage-ii-openai-published-weak"]},{"id":"eval-i-rsp-christiano-credit","statement":"Anthropic RSP v1.0 says its commitments are designed in the spirit of the Responsible Scaling Policy framework being developed by Paul Christiano and ARC Evals.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Scoped attribution: the policy names Christiano, with ARC Evals, as developing the RSP framework and says its commitments are designed in that framework’s spirit, alongside emerging government proposals. It does not say the commitments were derived from, adopted from or endorsed by Christiano; the paragraph’s explicit thanks go to ARC Evals only. Not sole authorship or proof that every commitment was implemented. Requalified on 2026-10-03 from “credits a framework”. Distinct from the record of the RSP announcement’s explicit credit to ARC Evals’ framework.","evidenceIds":["evidence-eval-i-rsp-christiano-credit"]},{"id":"provchange-fsf-v3-1-response-plan-distant-from-ccl","statement":"v3.1 states that if risks remain acceptable and model capabilities remain distant from a CCL, the response plan may include updating the alert threshold to ensure the safety buffer remains appropriate; v3.0 stated that where capabilities remain quite distant from a CCL, a response plan may involve adopting additional capability assessment processes to flag when heightened mitigations are required.","derivation":"inferred","resolution":"supported","inferenceRationale":"Premises: FSF · v3.1 states the provision as quoted in its provision record (PDF p. 6 (printed p. 6), §1.3.3 Risk Mitigation, second paragraph, fourth sentence); FSF · v3.0 states it as quoted in its provision record (PDF p. 6 (printed p. 6), §1.4 Response Plans and Mitigations, second paragraph, third sentence). The difference described is read directly from the two passages.","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-fsf-v3-1-response-plan-distant-from-ccl-later","evidence-provchange-fsf-v3-1-response-plan-distant-from-ccl-earlier"],"attributes":{"changeBasis":"document_comparison","provision":{"kind":"required_action","key":"fsf-response-plan-distant-from-ccl"}}},{"id":"prov-openai-group-pbc-cert-class-n-pre-approval","statement":"The certificate states that, so long as Class N Common Stock is issued, outstanding and held by OpenAI Foundation, the prior written approval of its holder is required for the corporation to amend Article III (Public Benefit); to agree to or complete a deemed liquidation event or a sale of material assets of the corporation and its subsidiaries taken as a whole; to issue more Class N stock; to create or issue other stock entitled to vote on directors, except Class B stock in acquisitions; or to modify the certificate or bylaws in a way that disproportionately and adversely affects the Class N holder.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"What the filed text says. The Delaware Attorney General’s statement records OpenAI’s representation of the same approval rights plus a sixth, over amendments to the PBC’s Corporate Governance Guidelines, “as further described in definitive documents”, which were not inspected (context below).","evidenceIds":["evidence-prov-openai-group-pbc-cert-class-n-pre-approval-1","evidence-prov-openai-group-pbc-cert-class-n-pre-approval-2"],"attributes":{"provision":{"kind":"decision_authority","key":"class-n-pre-approval","parties":[{"role":"decides","actor":"Holder of the Class N Common Stock"}]}}},{"id":"cm-seoul-mistral-ai","statement":"Mistral AI 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.","evidenceIds":["evidence-cm-seoul-mistral-ai-1"],"attributes":{"commitment":{"stage":"listed"}}},{"id":"prov-fasc-2024-threshold-input","statement":"Commitment II states that the thresholds should be defined with input from trusted actors, including the organisations’ respective home governments as appropriate.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"What the text says about consultation; it does not give any government a decision.","evidenceIds":["evidence-prov-fasc-2024-threshold-input-1"],"attributes":{"provision":{"kind":"decision_authority","key":"threshold-input","parties":[{"role":"consulted","actor":"Trusted actors, including home governments as appropriate"}]}}},{"id":"ts-a4-hanson-1996-09-03-contests","statement":"Hanson’s reply of 1996-09-03 to Bostrom concedes that one hostile super-AI far ahead of the rest would leave people at its mercy, but argues that any one AI probably won’t be too far ahead of any other, so they can police each other. Recorded as contesting the proposition “Fast or hard takeoff”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"It disputes the single-leader premise of Bostrom’s scenario, not concealment itself.","evidenceIds":["evidence-ts-a4-hanson-1996-09-03-contests"],"attributes":{"thought":{"roles":["contesting"],"register":"nonfiction","qualifier":"Disputes that any one AI would be far ahead of the rest.","engagement":{"kind":"contrary-only","rationale":"States the contrary expectation that no one AI will be far ahead of the others, without giving a reason against the premise that a single system could move quickly far beyond the rest."}}}},{"id":"lg-oai-simo-ceo-applications-snapshot-2025","statement":"OpenAI’s announcement of 2 September 2025 attributes a statement to Fidji Simo as CEO of Applications, OpenAI.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A dated title on a document; it does not establish when the role began.","evidenceIds":["lg-ev-oai-simo-ceo-applications-snapshot-2025"],"attributes":{"role":"CEO of Applications"}},{"id":"lg-deal-luan-openai-engineering-prior","statement":"David Luan wrote on 26 April 2022 of having led Engineering at OpenAI when it scaled up the Transformer into GPT-2 and GPT-3.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"First-person retrospective; it gives no formal title or dates, and no OpenAI source was retrieved. The GPT-2 and GPT-3 names do not date the role.","evidenceIds":["lg-ev-deal-luan-openai-engineering-prior"],"attributes":{"role":"led Engineering"}},{"id":"cmp-fsf-v3-1-d4-determination","statement":"FSF · v3.1 states a provision on the comparison question “Determination of capability or safeguard adequacy”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The 'appropriate governance function' is not named. The same function may also exempt low-risk external deployments from a critical capability assessment (see “Other exceptions or discretion”). What the edition states, quoted from the cited expression. It is not evidence of implementation, compliance or effectiveness, and a shared question does not make two editions’ provisions equivalent.","evidenceIds":["ev-cmp-fsf-v3-1-d4-determination-1"],"attributes":{"provision":{"kind":"decision_authority","key":"comparison-d4-determination","facet":{"id":"D4.determination","status":"stated"}}}},{"id":"cmp-ms-fgf-2025-d3-halt","statement":"Microsoft FGF · Feb 2025 states a provision on the comparison question “Pause, non-deployment or non-training”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The pause is tied to unmitigable risk rather than to reaching a level; high or critical models are 'subject to further review and safety and security mitigations prior to deployment' (PDF p. 7). What the edition states, quoted from the cited expression. It is not evidence of implementation, compliance or effectiveness, and a shared question does not make two editions’ provisions equivalent.","evidenceIds":["ev-cmp-ms-fgf-2025-d3-halt-1"],"attributes":{"provision":{"kind":"required_action","key":"comparison-d3-halt","facet":{"id":"D3.halt","status":"stated"}}}},{"id":"ap-robin-hanson-hanson-1994-if-uploads-come-first","statement":"“If Uploads Come First: The Crack of a Future Dawn” (1994) credits Robin Hanson as its author.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Credited authorship of this expression; it says nothing about priority for the ideas discussed.","evidenceIds":["evidence-ap-robin-hanson-hanson-1994-if-uploads-come-first"]},{"id":"cmp-rsp-v3-0-d5-cadence","statement":"RSP · v3.0 states a provision on the comparison question “Evaluation timing”, in a weaker form than the question’s criterion.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Stated, partly: the passage has what the question’s criterion asks for, but in a weaker form than the criterion’s own words. The edition sets how often Risk Reports, which include capability and alignment evaluation results, are published, not a schedule for running evaluations. It adds interim discussions when a publicly deployed model is significantly more capable than the models in the latest Risk Report, and no later than 30 days after Anthropic determines that an internally deployed model is in scope. No compute- or capability-interval trigger is stated. What the edition states, paraphrased from the cited expression with its page. It is not evidence of implementation, compliance or effectiveness, and a shared question does not make two editions’ provisions equivalent.","evidenceIds":["ev-cmp-rsp-v3-0-d5-cadence-1"],"attributes":{"provision":{"kind":"evaluation_timing","key":"comparison-d5-cadence","facet":{"id":"D5.cadence","status":"partly"}}}},{"id":"cmp-fsf-2024-d1-cbrn","statement":"FSF · 2024 states a provision on the comparison question “Chemical, biological, radiological or nuclear weapons”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Only the biological domain (named “Biosecurity”) has CCLs in this edition: Bio amateur enablement level 1 and Bio expert enablement level 1. Chemical, radiological and nuclear risks appear only in Future work, as possible candidates for inclusion that existing model evaluations outside the Framework already cover. What the edition states, paraphrased from the cited expression with its page. It is not evidence of implementation, compliance or effectiveness, and a shared question does not make two editions’ provisions equivalent.","evidenceIds":["ev-cmp-fsf-2024-d1-cbrn-1","ev-cmp-fsf-2024-d1-cbrn-2"],"attributes":{"provision":{"kind":"risk_domain","key":"comparison-d1-cbrn","facet":{"id":"D1.cbrn","status":"stated"}}}},{"id":"c2-governance-aiia-aisc-signatory-ant-group","statement":"Ant Group is listed, as 蚂蚁科技集团股份有限公司, among the 17 companies on the initial signatory list (首批签署企业名单) of the AIIA AI Safety Commitments posted by CAICT on 24 December 2024.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The list names the legal entity 蚂蚁科技集团股份有限公司. The record maps it to the group-level entity Ant Group, which carries that name as an alias; the mapping rests on the company name, and no claim is made about group structure or about which group company is bound. A listing is the publisher’s statement, dated by the post; it is not a signature date and not evidence that the company complied with or implemented the commitments. A later total of 22 signatories is reported, but no later list was retrieved.","evidenceIds":["c2-ev-governance-aiia-aisc-signatory-ant-group-1","c2-ev-governance-aiia-aisc-signatory-ant-group-2"],"attributes":{"commitment":{"stage":"listed"}}},{"id":"fl-us-ai-action-plan-caisi-frontier-evaluations","statement":"America’s AI Action Plan recommends that CAISI at the Department of Commerce lead evaluations of frontier AI systems for national security risks in partnership with frontier AI developers.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A recommended policy action in a plan, not a legal duty; it names no developer and records no evaluation.","evidenceIds":["evidence-fl-us-ai-action-plan-caisi-frontier-evaluations-1"],"attributes":{"provision":{"kind":"required_action","key":"us-ai-action-plan-caisi-frontier-evaluations"}}},{"id":"cpo-dc-uk-aigz-five-designated","statement":"The UK government's January 2026 progress update stated that it had designated 5 AI Growth Zones across Great Britain, including 2 in Wales and one in Scotland, generating £28.2 billion in investment.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A government progress claim; \"generating £28.2 billion in investment\" is the government's figure and is not broken down by zone in the passage. Designation is not construction.","evidenceIds":["evidence-cpo-dc-uk-aigz-five-designated-1"]},{"id":"ab-gov-us-action-plan-biosecurity-section","statement":"America's AI Action Plan states, in its \"Invest in Biosecurity\" section, that while AI will unlock potential in biology it could create new pathways for malicious actors to synthesize harmful pathogens and other biomolecules, and that the solution is a multi-tiered approach to screen for malicious actors together with new tools and infrastructure for more effective screening.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The plan's own framing of a risk and of its preferred approach. It is not a finding about any AI system's capability, and the plan states no evidence for the risk in this section. The passage is cited at the plan's own generality; no agent, method or route is named in the source passage or in this record.","evidenceIds":["evidence-ab-gov-us-action-plan-biosecurity-section-1"],"attributes":{"provision":{"kind":"risk_domain","key":"us-ai-action-plan-biosecurity-framing"}}},{"id":"ab-ev-claude37-uplift-study-result","statement":"The Claude 3.7 Sonnet System Card states that in one uplift study models provided some amount of uplift to novices compared with participants who did not have access to a model, and that even the highest scoring plan from a participant with model access still included critical mistakes that would lead to failure in the real world.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"One study as the developer reports it, for the models and participants it describes. The record does not say how large any uplift was, does not generalise it to other systems or settings, and carries the card's own statement that plans still failed on critical points. The study's tasks and the card's description of where performance improved are technical content and are excluded; only the stated comparative outcome is recorded.","evidenceIds":["evidence-ab-ev-claude37-uplift-study-result-1","evidence-ab-ev-claude37-uplift-study-result-2","evidence-ab-ev-claude37-uplift-study-result-3"],"attributes":{"evaluatedSystem":"Claude 3.7 Sonnet and, where the card says so, a helpful-only version of it","evaluation":{"what":"Whether access to a model increased what participants could produce on tasks proxying real-world CBRN tasks","how":"A human-participant uplift study with model-access and no-model-access conditions, scored against the study's own criteria","result":"Some amount of uplift to novices compared with participants without model access; the highest scoring plan with model access still included critical mistakes that would lead to failure in the real world","caveats":"The card states that results differed across its evaluation strategies, with some showing uplift and others not, and that expert-level evaluation performance may not directly translate to real-world capabilities."}}}],"sources":[{"id":"source-legg-deepmind-2014","title":"From academia to industry: The story of Google DeepMind","authors":["Shane Legg"],"publisher":"Schloss Dagstuhl – Leibniz-Zentrum für Informatik","publicationDate":"2014-10-08","version":"ICCSW 2014, OASIcs volume 43, page 1. Founder-authored proceedings abstract; DOI 10.4230/OASIcs.ICCSW.2014.1.","canonicalUrl":"https://drops.dagstuhl.de/entities/document/10.4230/OASIcs.ICCSW.2014.1","evidenceUrl":"https://drops.dagstuhl.de/entities/document/10.4230/OASIcs.ICCSW.2014.1","language":"en","archivedUrl":"https://web.archive.org/web/20260420154356/https://drops.dagstuhl.de/entities/document/10.4230/OASIcs.ICCSW.2014.1","archivedAt":"2026-04-20","accessedAt":"2026-09-18","reuse":{"attribution":"Shane Legg, From academia to industry: The story of Google DeepMind (2014-10-08).","licence":"Creative Commons Attribution 3.0 Unported","note":"Bibliographic metadata and brief attributed paraphrases or excerpts only. Source roles are relative to each claim."}},{"id":"source-rsp-1-0","title":"Responsible Scaling Policy — v1.0","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":"2023-09-19","version":"Responsible Scaling Policy v1.0; effective date 2023-09-19.","versionDate":"2023-09-19","canonicalUrl":"https://www-cdn.anthropic.com/1adf000c8f675958c2ee23805d91aaade1cd4613/responsible-scaling-policy.pdf","evidenceUrl":"https://www-cdn.anthropic.com/1adf000c8f675958c2ee23805d91aaade1cd4613/responsible-scaling-policy.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20230919161730/https://www-files.anthropic.com/production/files/responsible-scaling-policy-1.0.pdf","archivedAt":"2023-09-19","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — 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-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-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-fsf-3-0","title":"Frontier Safety Framework — v3.0","authors":["Google DeepMind"],"publisher":"Google DeepMind","publicationDate":"2025-09-22","version":"Frontier Safety Framework v3.0; publication date 2025-09-22.","versionDate":"2025-09-22","canonicalUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3.pdf","evidenceUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250922134959/https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3.pdf","archivedAt":"2025-09-22","accessedAt":"2026-09-18","reuse":{"attribution":"Google DeepMind, Frontier Safety Framework — 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-fsf-3-1","title":"Frontier Safety Framework — v3.1","authors":["Google DeepMind"],"publisher":"Google DeepMind","publicationDate":"2026-04-17","version":"Frontier Safety Framework v3.1; publication date 2026-04-17.","versionDate":"2026-04-17","canonicalUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf","evidenceUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260423021420/https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf","archivedAt":"2026-04-23","accessedAt":"2026-09-18","reuse":{"attribution":"Google DeepMind, Frontier Safety Framework — 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-ssi-launch-reuters-2024","title":"Former OpenAI chief scientist to start a new AI company","authors":["Reuters"],"publisher":"Reuters","publicationDate":"2024-06-19","version":"Contemporaneous Reuters report inspected in its public syndication. The original SSI founding page has changed; web archive retrieval failed. Secondary reporting is labelled as such. Copy read on Investing.com.","canonicalUrl":"https://www.investing.com/news/stock-market-news/former-openai-chief-scientist-to-start-a-new-ai-company-3489233","evidenceUrl":"https://www.investing.com/news/stock-market-news/former-openai-chief-scientist-to-start-a-new-ai-company-3489233","language":"en","archivedUrl":"https://web.archive.org/web/20240811052921/https://www.investing.com/news/stock-market-news/former-openai-chief-scientist-to-start-a-new-ai-company-3489233","archivedAt":"2024-08-11","accessedAt":"2026-09-19","reuse":{"attribution":"Reuters, Former OpenAI chief scientist to start a new AI company (2024-06-19)","licence":"No open redistribution licence asserted","note":"Bibliographic metadata and brief attributed paraphrases only; no full source or personal contact details redistributed."}},{"id":"source-weak-strong-announcement","title":"Weak-to-strong generalization","authors":["Collin Burns","Jan Leike","Leopold Aschenbrenner","Jeffrey Wu","Pavel Izmailov","Leo Gao","Bowen Baker","Jan Hendrik Kirchner"],"publisher":"OpenAI","publicationDate":"2023-12-14","version":"Mutable publisher announcement displaying 2023-12-14; paper link identifies the work but does not pin the hosted PDF revision.","canonicalUrl":"https://openai.com/index/weak-to-strong-generalization/","evidenceUrl":"https://openai.com/index/weak-to-strong-generalization/","language":"en","archivedUrl":"https://web.archive.org/web/20260917090610/https://openai.com/index/weak-to-strong-generalization/","archivedAt":"2026-09-17","accessedAt":"2026-09-25","reuse":{"attribution":"Collin Burns; Jan Leike; Leopold Aschenbrenner; Jeffrey Wu; Pavel Izmailov; Leo Gao; Bowen Baker; Jan Hendrik Kirchner, Weak-to-strong generalization.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata and brief attributed paraphrases only; source publications and personal contact details are not redistributed."}},{"id":"source-ms-fgf-2025","title":"Frontier Governance Framework — Version 1 (February 2025)","authors":["Microsoft"],"publisher":"Microsoft","publicationDate":"2025-02","version":"Frontier Governance Framework, Version 1 (February 2025); its change log reads “8 February 2025 – First version”. The file served at the recorded address (last modified 2025-02-12) is byte-identical to the Internet Archive capture of 2025-02-26, the earliest located.","versionDate":"2025-02-08","canonicalUrl":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/final/en-us/microsoft-brand/documents/Microsoft-Frontier-Governance-Framework.pdf","evidenceUrl":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/final/en-us/microsoft-brand/documents/Microsoft-Frontier-Governance-Framework.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250226161803/https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/final/en-us/microsoft-brand/documents/Microsoft-Frontier-Governance-Framework.pdf","archivedAt":"2025-02-26","accessedAt":"2026-10-03","reuse":{"attribution":"Microsoft, Frontier Governance Framework, Version 1 (February 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-openai-group-pbc-certificate","title":"Certificate of Incorporation of OpenAI Group PBC","authors":["OpenAI Group PBC"],"publisher":"OpenAI Group PBC (filed with the Delaware Division of Corporations)","publicationDate":"2025-10-28","version":"Filed 2025-10-28 at 08:07 AM according to the Division of Corporations stamp (file number 10381551); 30-page scan with an OCR text layer, downloaded from DocumentCloud on 2026-10-04 and byte-identical (SHA-1) to the Internet Archive captures from 2025-10-29. Quotes were read from the page images because the OCR layer contains recognition errors.","canonicalUrl":"https://s3.documentcloud.org/documents/26205026/openai-pbc-articles-of-incorporation.pdf","evidenceUrl":"https://s3.documentcloud.org/documents/26205026/openai-pbc-articles-of-incorporation.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20251029204930/https://s3.documentcloud.org/documents/26205026/openai-pbc-articles-of-incorporation.pdf","archivedAt":"2025-10-29","accessedAt":"2026-10-04","reuse":{"attribution":"OpenAI Group PBC, Certificate of Incorporation (filed 28 October 2025), copy hosted by DocumentCloud.","licence":"Public filing; 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-delaware-ag-openai-2025","title":"Letter re: OpenAI, Inc.’s Corporate Restructuring (statement of no objection)","authors":["Delaware Department of Justice, Office of the Attorney General"],"publisher":"Delaware Department of Justice (Office of the Attorney General)","publicationDate":"2025-10-28","version":"Letter dated October 28, 2025, to OpenAI, Inc.; PDF inspected 2026-10-04, byte-identical to the Internet Archive capture of 2025-10-31. The earlier capture of 2025-10-28 differs only in also giving the recipient’s e-mail address.","canonicalUrl":"https://news.delaware.gov/files/2025/10/2025-10-28-OpenAI-DEAG-Statement-of-Nonobjection.pdf","evidenceUrl":"https://news.delaware.gov/files/2025/10/2025-10-28-OpenAI-DEAG-Statement-of-Nonobjection.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20251031110818/https://news.delaware.gov/files/2025/10/2025-10-28-OpenAI-DEAG-Statement-of-Nonobjection.pdf","archivedAt":"2025-10-31","accessedAt":"2026-10-04","reuse":{"attribution":"Delaware Department of Justice, letter to OpenAI, Inc. re: OpenAI, Inc.’s Corporate Restructuring (28 October 2025).","licence":"Public government document; 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-1996-09-03-hanson","title":"Re: Thinking about the future... (Extropians mailing list)","authors":["Robin Hanson"],"publisher":"Extropians mailing list","publicationDate":"1996-09-03","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.96/0731.html","evidenceUrl":"http://extropians.weidai.com/extropians.96/0731.html","language":"en","archivedUrl":"https://web.archive.org/web/20150427164816/http://extropians.weidai.com/extropians.96/0731.html","archivedAt":"2015-04-27","accessedAt":"2026-10-06","reuse":{"attribution":"Robin Hanson, \"Re: Thinking about the future...\", Extropians mailing list, 1996-09-03.","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-oai-statsig-2025","title":"Vijaye Raji to become CTO of Applications with acquisition of Statsig","authors":["OpenAI"],"publisher":"OpenAI","publicationDate":"2025-09-02","version":"Dated page, retrieved on 2026-10-15 (SHA-256 9846c1c849eee13f017372ee5c60964fc140b203b68d966d47430d4c74dc7b11) through the third-party reader relay r.jina.ai, which renders the publisher’s current page (the publisher refused direct retrieval); not an archive copy.","canonicalUrl":"https://openai.com/index/vijaye-raji-to-become-cto-of-applications-with-acquisition-of-statsig/","evidenceUrl":"https://openai.com/index/vijaye-raji-to-become-cto-of-applications-with-acquisition-of-statsig/","language":"en","archivedUrl":"https://web.archive.org/web/20260815103133/https://openai.com/index/vijaye-raji-to-become-cto-of-applications-with-acquisition-of-statsig/","archivedAt":"2026-08-15","accessedAt":"2026-10-15","reuse":{"attribution":"OpenAI, “Vijaye Raji to become CTO of Applications with acquisition of Statsig”, 2025-09-02.","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":"lg-src-deal-adept-intro-2022","title":"Introducing Adept","authors":["David Luan"],"publisher":"Adept","publicationDate":"2022-04-26","version":"Dated post, retrieved on 2026-10-15 (SHA-256 e669456e92b54c5ee26d63eda9bce65bff8dc64ced799c8b3a45fd748ae20b3c) through the third-party reader relay r.jina.ai, which renders the publisher’s current page (the publisher refused direct retrieval); not an archive copy; a fresh retrieval the same day got a block page. The current page is a later expression of the post. Archived copy: the Wayback Machine capture of 2022-04-26 16:02:16 UTC of the original address (/post/introducing-adept), titled “Introducing Adept AI Labs”, read by a cross-family machine check on 2026-10-26; no byte equivalence between the capture and the 2026 retrieval is claimed.","canonicalUrl":"https://www.adept.ai/blog/introducing-adept","evidenceUrl":"https://www.adept.ai/blog/introducing-adept","language":"en","archivedUrl":"https://web.archive.org/web/20220426160216/https://www.adept.ai/post/introducing-adept","archivedAt":"2022-04-26","accessedAt":"2026-10-15","reuse":{"attribution":"David Luan, “Introducing Adept”, 2022-04-26.","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":"c2-src-caict-wx-aisc-2024","title":"守护AI安全，共建行业自律典范——首批17家企业签署《人工智能安全承诺》","authors":["中国信通院CAICT"],"publisher":"中国信息通信研究院 (CAICT), WeChat official account 中国信通院CAICT","publicationDate":"2024-12-24","version":"Dated post, retrieved on 2026-10-27 (SHA-256 4a3b099c801423b12830598c841921a74d01d4d9d3f69379669aaf17e6c38d78) directly from the publisher. The commitment text (Chinese and English) and the initial signatory list are a single embedded image (1000 × 9898 pixels, SHA-256 7e9145350d1b882d22ab3f7fc4d4c0226f06194f94264d92291ee065fbf4aa27), not in the page’s text layer; quotations from it are transcriptions. The cross-family check found the same image in an Internet Archive capture of 4 March 2026 with the same SHA-256. Its wording can change; earlier wording is not authenticated. The cross-family check (2026-10-27) located the Internet Archive capture linked here; a capture shows only what the archive retrieved that day. English title (machine translation): “Safeguarding AI safety, building a model of industry self-discipline: initial 17 companies sign the AI Safety Commitments”.","canonicalUrl":"https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng","evidenceUrl":"https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng","language":"zh-Hans","archivedUrl":"https://web.archive.org/web/20260724145912/https://mp.weixin.qq.com/s/s-XFKQCWhu0uye4opgb3Ng","archivedAt":"2026-07-24","accessedAt":"2026-10-27","reuse":{"attribution":"中国信通院CAICT, “守护AI安全，共建行业自律典范——首批17家企业签署《人工智能安全承诺》”, 2024-12-24.","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":"c2-src-itcloudbd-aisc-2024","title":"守护AI安全，共建行业自律典范 首批17家企业签署《人工智能安全承诺》","authors":["科技云报到"],"publisher":"科技云报到 (itcloudbd.com)","publicationDate":"2024-12-27","version":"Dated page, retrieved on 2026-10-27 (SHA-256 f4d6790217cccacfc6fea787516f4538abb1d4c1bf03b65d3f5123407c0d72f4) directly from the publisher. Tech-media article dated 2024-12-27, read on 2026-10-27. It reproduces the CAICT post's wording and gives the commitments and signatory names as text; its commitment-two wording (红队测试) matches the image in the CAICT post. Its wording can change; earlier wording is not authenticated. The cross-family check (2026-10-27) located the Internet Archive capture linked here; a capture shows only what the archive retrieved that day. English title (machine translation): “First 17 companies sign the AI Safety Commitments”. A reporting outlet, cited beside the primary text.","canonicalUrl":"https://itcloudbd.com/rengongzhineng/13352.html","evidenceUrl":"https://itcloudbd.com/rengongzhineng/13352.html","language":"zh-Hans","archivedUrl":"https://web.archive.org/web/20260712204752/https://itcloudbd.com/rengongzhineng/13352.html","archivedAt":"2026-07-12","accessedAt":"2026-10-27","reuse":{"attribution":"科技云报到, “守护AI安全，共建行业自律典范 首批17家企业签署《人工智能安全承诺》”, 2024-12-27.","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":"source-us-ai-action-plan-2025","title":"Winning the Race: America’s AI Action Plan (July 2025)","authors":["The White House"],"publisher":"The White House","publicationDate":"2025-07","version":"Retrieved on 2026-10-27 (SHA-256 91d56ab3ddefe43f395bde31e1c3923fb37f47091275aaf2a0d7da500ff547f4). PDF as served at whitehouse.gov on 2026-10-27; cover dated \"JULY 2025\". The PDF text still uses \"AI\".","canonicalUrl":"https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf","evidenceUrl":"https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf","language":"en","accessedAt":"2026-10-27","reuse":{"attribution":"The White House, “Winning the Race: America’s AI Action Plan (July 2025)”, 2025-07.","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-govuk-aiop-one-year-2026","title":"AI Opportunities Action Plan: One Year On","authors":["Department for Science, Innovation and Technology (GOV.UK)"],"publisher":"Department for Science, Innovation and Technology (GOV.UK)","publicationDate":"2026-01-29","version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 698f8b2bfee393478e14da1a2eb70881c57becbe291b1111246fb609b2fc18bc). GOV.UK policy paper (HTML) as served on 2026-10-28; \"Published 29 January 2026\". Retrieved again at admission on 2026-10-30 (SHA-256 923323c25077abd19e4a15343d5289d6c55e008cbf5eb11fbd7bdcbfcc571837).","canonicalUrl":"https://www.gov.uk/government/publications/ai-opportunities-action-plan-one-year-on/ai-opportunities-action-plan-one-year-on","evidenceUrl":"https://www.gov.uk/government/publications/ai-opportunities-action-plan-one-year-on/ai-opportunities-action-plan-one-year-on","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"Department for Science, Innovation and Technology (GOV.UK), “AI Opportunities Action Plan: One Year On”, 2026-01-29.","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-ab-claude37-card","title":"Claude 3.7 Sonnet System Card","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 82947a46fc2d1b9a1d69501ef973fc3c5b328b703b9a17bd8a143c2672f8a1a6). 42-page PDF as served on 2026-10-28; the request to assets.anthropic.com redirected to https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf. The document itself states no publication date; its references are dated February 2025. Retrieved again at admission on 2026-10-30 (SHA-256 82947a46fc2d1b9a1d69501ef973fc3c5b328b703b9a17bd8a143c2672f8a1a6, the same bytes).","canonicalUrl":"https://assets.anthropic.com/m/785e231869ea8b3b/original/claude-3-7-sonnet-system-card.pdf","evidenceUrl":"https://assets.anthropic.com/m/785e231869ea8b3b/original/claude-3-7-sonnet-system-card.pdf","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"Anthropic, “Claude 3.7 Sonnet System Card”.","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-google-acquired-deepmind-2014","sourceRole":"primary","support":"supports","locator":"Abstract, second paragraph, final sentence","passage":"Legg’s 2014 abstract describes Google’s acquisition earlier that year.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-18","note":"Relevant source passage inspected by an automated research assistant. Human source review is pending."},{"id":"evidence-daniel-gross-ssi-cofounder-2024","sourceRole":"secondary","support":"supports","locator":"Body, company launch and co-founders paragraphs","passage":"Reuters names Daniel Gross among SSI’s co-founders.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-19","note":"Source passage independently inspected by an automated research assistant for this scoped record. Human source review pending."},{"id":"evidence-lineage-ii-openai-published-weak","sourceRole":"primary","support":"supports","locator":"The superalignment problem, second paragraph; Read paper link","passage":"OpenAI announces release of the team’s first paper and links the Weak-to-Strong Generalization work.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-25","note":"Direct source inspection; machine check only. Publisher’s attribution of programme output. Mutable expression displaying 2023-12-14, not authenticated wording from that day; not independent verification of results or derivation from Good/MIRI."},{"id":"evidence-eval-i-rsp-christiano-credit","sourceRole":"primary","support":"supports","locator":"PDF p. 1, introduction, fourth paragraph, first sentence","passage":"Our commitments are designed in the spirit of the Responsible Scaling Policy (RSP) framework being developed by Paul Christiano and ARC Evals, as well as emerging government policy proposals on responsible AI development in the UK, EU, and US.","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 2023-09-19 (earlier www-files.anthropic.com address); the whole sentence is quoted so its second limb is not lost. This is not human review or independent reproduction of the reported activity."},{"id":"evidence-provchange-fsf-v3-1-response-plan-distant-from-ccl-later","sourceRole":"primary","support":"supports","locator":"PDF p. 6 (printed p. 6), §1.3.3 Risk Mitigation, second paragraph, fourth sentence","passage":"FSF v3.1 states that if Google DeepMind assesses that risks stemming from the model remain acceptable and model capabilities remain distant from a CCL, the response plan may include updating the alert threshold to ensure the safety buffer remains appropriate.","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-fsf-v3-1-response-plan-distant-from-ccl-earlier","sourceRole":"primary","support":"supports","locator":"PDF p. 6 (printed p. 6), §1.4 Response Plans and Mitigations, second paragraph, third sentence","passage":"FSF v3.0 states that where model capabilities remain quite distant from a CCL, a response plan may involve the adoption of additional capability assessment processes to flag when heightened mitigations are required.","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-prov-openai-group-pbc-cert-class-n-pre-approval-1","sourceRole":"primary","support":"supports","locator":"Section IV.2(c) “Special Class N Pre-Approval Rights” (page 5)","passage":"the prior written approval of the holder of the outstanding shares of Class N Common Stock shall be required in order for the Corporation to: (i) amend any provision of Article III (Public Benefit)","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-04","note":"Verbatim quote transcribed from the page image by the integrating session on 2026-10-04 and compared with the scan; the copy has no reliable text layer, so the match is visual, not mechanical. 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-prov-openai-group-pbc-cert-class-n-pre-approval-2","sourceRole":"primary","support":"context","locator":"Representation 5 (pages 2–3)","passage":"The letter records OpenAI’s representation that, as long as the nonprofit holds Class N Common Stock, its prior written approval will be required, as further described in definitive documents, for six actions: the five in the certificate and amendments to the PBC’s Corporate Governance Guidelines.","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 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-cm-seoul-mistral-ai-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 “Mistral AI”.","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-prov-fasc-2024-threshold-input-1","sourceRole":"primary","support":"supports","locator":"Outcome 1, commitment II, third sentence","passage":"defined with input from trusted actors, including organisations’ respective home governments as appropriate","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 2024-05-21. The passage was first proposed and machine-verified in the week 4b research pass. Human editorial review is pending."},{"id":"evidence-ts-a4-hanson-1996-09-03-contests","sourceRole":"primary","support":"supports","locator":"Message body, unquoted reply to the quoted Bostrom passage","passage":"any one AI probably won't be too far ahead of any other AI","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-28","note":"Checked against the archive HTML on 2026-10-06; shortened from the candidate search’s quotation to fit the limit for a 169-word message. On 2026-10-30, by the maintainer’s decision of that day, the quotation was corrected to the archive capture of 2015-04-27 (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-oai-simo-ceo-applications-snapshot-2025","sourceRole":"primary","support":"supports","locator":"Attribution of the quotation following the second paragraph","passage":"Fidji Simo, CEO of Applications, OpenAI","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (6 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read through the third-party reader relay r.jina.ai, which renders the publisher’s current page (the publisher refused direct retrieval); not an archive copy. 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":"reader_relay"},{"id":"lg-ev-deal-luan-openai-engineering-prior","sourceRole":"primary","support":"supports","locator":"Paragraph beginning “The Transformer”","passage":"When I led Engineering at OpenAI, we scaled up the Transformer into GPT-2 and GPT-3","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-26","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (15 words) matched the text read through the third-party reader relay r.jina.ai, which renders the publisher’s current page (the publisher refused direct retrieval); not an archive copy on 2026-10-15, but a fresh retrieval the same day got a block page, so the record was held back at the genealogy admission of 2026-10-17. The passage now rests on the cross-family machine check’s read of the Wayback Machine capture of 2022-04-26 16:02:16 UTC, made on 2026-10-26 (research/archive-checks-2026-10-26/codex-report.md, item 16), which found the record supported; the capture could not be re-fetched when the record was admitted (2026-10-26). In the capture the post is titled “Introducing Adept AI Labs”, at /post/introducing-adept; the locator names the paragraph as it opens there. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. Admitted by the maintainer’s decision of 2026-10-26. No human editorial review."},{"id":"ev-cmp-fsf-v3-1-d4-determination-1","sourceRole":"primary","support":"supports","locator":"PDF p. 10 (printed p. 10), §2.1.2 Deployment Mitigations, numbered step 2, first sentence","passage":"Pre-deployment review of residual risk assessment: external deployments of a model take place only after the appropriate governance function determines the residual risk to be acceptable (including a safety case where a CCL has been reached).","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-27","note":"Existing dataset passage for this edition, which contains the comparison’s shorter quotation, first checked on 2026-10-03. The comparison’s quotation was checked by an automated research assistant on 2026-10-27 against the publisher copy, whose SHA-256 matches the copy retrieved on 2026-10-15. The sentence continues '(including a safety case where a CCL has been reached)'. Human editorial review is pending."},{"id":"ev-cmp-ms-fgf-2025-d3-halt-1","sourceRole":"primary","support":"supports","locator":"PDF p. 9 (printed p. 8), §4 Mitigations, Safety mitigations, paragraph 'Post-mitigation capability assessment and safety buffer', second sentence","passage":"If, during the implementation of this framework, we identify a risk we cannot sufficiently mitigate, we will pause development and deployment until the point at which mitigation practices evolve to meet the risk.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-27","note":"Existing dataset passage for this edition, which contains the comparison’s shorter quotation, first checked on 2026-10-03. The comparison’s quotation was checked by an automated research assistant on 2026-10-27 against the publisher copy, whose SHA-256 matches the copy retrieved on 2026-10-15. Human editorial review is pending."},{"id":"evidence-ap-robin-hanson-hanson-1994-if-uploads-come-first","sourceRole":"primary","support":"supports","locator":"PDF page 8 (printed page 10), byline beneath the article title; also in the contents on PDF page 1","passage":"Credited as “Robin Hanson” (pDF page 8 (printed page 10), byline beneath the article title; also in the contents on PDF page 1).","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-27","note":"Passage and locator as inspected by a reviewer from a different model family on 2026-10-27 in a direct retrieval of the issue scan; the preparer’s retrieval of 2026-10-27 had failed, so the check’s reading replaces a restatement of the source record’s author list; the scan has no text layer, so the page was checked visually. Given as a labelled paraphrase because the source records no document length against which a quotation can be checked. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"ev-cmp-rsp-v3-0-d5-cadence-1","sourceRole":"primary","support":"supports","locator":"PDF p. 10, §3.1 Scope and Timing, paragraph 'Timing'","passage":"The edition says a Risk Report will be published every three to six months.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-27","note":"Explicit paraphrase, written on 2026-10-27 by an automated research assistant after re-reading the passage in the publisher copy, whose SHA-256 matches the copy retrieved on 2026-10-15. The comparison quoted the passage; it is given here as a paraphrase to keep the edition within the quotation limit. Human editorial review is pending."},{"id":"ev-cmp-fsf-2024-d1-cbrn-1","sourceRole":"primary","support":"supports","locator":"PDF p. 1 (printed p. 1), opening section, second paragraph, third sentence","passage":"The edition says its first CCLs cover four domains: Autonomy, Biosecurity, Cybersecurity and “Machine Learning R&D”.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-27","note":"Explicit paraphrase, written on 2026-10-27 by an automated research assistant after re-reading the passage in the publisher copy, whose SHA-256 matches the copy retrieved on 2026-10-15. The comparison quoted the passage; it is given here as a paraphrase to keep the edition within the quotation limit. Human editorial review is pending."},{"id":"ev-cmp-fsf-2024-d1-cbrn-2","sourceRole":"primary","support":"supports","locator":"PDF p. 6 (printed p. 6), Future work, bullet 'Updated set of risks and mitigations', second sub-bullet","passage":"Under Future work, the edition says Google DeepMind's existing model evaluations, which it is already implementing, currently cover powerful capabilities in the chemical, radiological and nuclear domains, and that these domains are possible candidates for inclusion in the Framework.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-27","note":"Explicit paraphrase, written on 2026-10-27 by an automated research assistant after re-reading the passage in the publisher copy, whose SHA-256 matches the copy retrieved on 2026-10-15. The comparison quoted the passage; it is given here as a paraphrase to keep the edition within the quotation limit. Human editorial review is pending."},{"id":"c2-ev-governance-aiia-aisc-signatory-ant-group-1","sourceRole":"primary","support":"supports","locator":"Embedded image, list 首批签署《人工智能安全承诺》企业名单 (transcribed from the image; not in the page’s text layer)","passage":"Transcription from the image: the list of initial signatories (首批签署《人工智能安全承诺》企业名单, 17 names, ordered by initial letter) includes 蚂蚁科技集团股份有限公司.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-27","note":"Candidate pack passage, Chinese frontier-AI institutions, second collection (research/china-labs-2). Transcription of a name that appears only in the image embedded in the post (the list of initial signatories); it is not in the page’s text layer and was not machine-matched. The drafting pass and the same-family audit read the list visually, and the cross-family check (2026-10-27) read all 17 names and agreed. The paraphrase is Selekoa’s; the listed name is transcribed as printed. No English rendering is given; no human translation review. A cross-family machine check (2026-10-27, item K011) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"c2-ev-governance-aiia-aisc-signatory-ant-group-2","sourceRole":"secondary","support":"supports","locator":"Final paragraph","passage":"智谱华章、第四范式、华为、科大讯飞、蚂蚁集团、360","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-27","note":"Candidate pack passage, Chinese frontier-AI institutions, second collection (research/china-labs-2). The quotation (19 words, each CJK character counted as one) matched the retrieved text on 2026-10-27 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. English machine translation, not verified by a person: “Zhipu Huazhang, 4Paradigm, Huawei, iFlytek, Ant Group, 360”. A cross-family machine check (2026-10-27, item K011) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-fl-us-ai-action-plan-caisi-frontier-evaluations-1","sourceRole":"primary","support":"supports","locator":"Pillar III, “Ensure that the U.S. Government is at the Forefront of Evaluating National Security Risks in Frontier Models”, Recommended Policy Actions, PDF p. 25 (printed p. 22)","passage":"Evaluate frontier AI systems for national security risks in partnership with frontier AI developers, led by CAISI at DOC in collaboration with other agencies with relevant expertise in CBRNE and cyber risks.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-28","note":"Candidate pack passage, frameworks to law (research/frameworks-to-law). The quotation (32 words) matched the text retrieved for the candidate pack on 2026-10-27 and again at admission on 2026-10-28, in a fresh retrieval whose SHA-256 equals the pack’s. Read directly from www.whitehouse.gov. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. Locator corrected from Pillar II to Pillar III (cross-family check, 2026-10-28, item 60). A cross-family machine check (2026-10-28, item 60) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-cpo-dc-uk-aigz-five-designated-1","sourceRole":"primary","support":"supports","locator":"section 1","passage":"The government designated 5 AI Growth Zones across Great Britain, including 2 in Wales and one in Scotland – generating £28.2 billion in investment","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, compute, power and orbit (research/compute-power-orbit). The quotation (23 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 (SHA-256 923323c25077abd19e4a15343d5289d6c55e008cbf5eb11fbd7bdcbfcc571837; the bytes differ from the pack’s retrieval). 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-ab-gov-us-action-plan-biosecurity-section-1","sourceRole":"primary","support":"supports","locator":"Pillar III, \"Invest in Biosecurity\", opening paragraph, PDF p. 26 (printed p. 23)","passage":"it could create new pathways for malicious actors to synthesize harmful pathogens and other biomolecules","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). 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. The paragraph continues that the solution is a multi-tiered approach designed to screen for malicious actors, along with new tools and infrastructure for more effective screening (not quoted: the PDF text layer breaks \"multi-tiered\" across a line). No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-uplift-study-result-1","sourceRole":"primary","support":"supports","locator":"PDF p. 23, §7.1 CBRN Evaluations, 'Results' paragraph","passage":"in one uplift study, we found that models provided some amount of uplift to novices compared to other participants who did not have access to a model","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (27 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 50) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-uplift-study-result-2","sourceRole":"primary","support":"context","locator":"PDF p. 23, §7.1, 'Results' paragraph, following sentence","passage":"even the highest scoring plan from the participant who had access to a model still included critical mistakes that would lead to failure in the \"real world\".","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (27 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 50) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-uplift-study-result-3","sourceRole":"primary","support":"context","locator":"PDF p. 23, §7.1, 'Results' paragraph","passage":"how the model performs on expert-level evaluations may not directly translate to real-world capabilities","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (14 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 card's own limit on reading its evaluation results. A cross-family machine check (2026-10-30, item 50) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"}],"at":{"entities":[0,9,21,22,39,45,51,52,75,81,103,106,201,223,233,243,252,282,290,339,657,670,884,936,937,1041,1236],"claims":[51,175,242,259,594,646,709,722,867,1801,2041,2116,2159,2246,2536,2589,2751,2990,3163,3343,3408],"sources":[25,35,39,44,46,47,80,106,183,215,216,239,284,291,764,772,778,787,877,990,1081],"evidence":[53,194,263,280,667,668,779,780,888,905,1081,2317,2639,2744,2803,2893,3187,3251,3252,3467,3468,3876,4216,4592,4690,4691,4692]},"version":"1lg47t7"}