{"entities":[{"id":"microsoft","name":"Microsoft"},{"id":"eliezer-yudkowsky","name":"Eliezer Yudkowsky"},{"id":"fsf-2024","name":"Frontier Safety Framework (2024)"},{"id":"mustafa-suleyman","name":"Mustafa Suleyman"},{"id":"google","name":"Google"},{"id":"rsp-v2-2","name":"Anthropic Responsible Scaling Policy — v2.2"},{"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":"summarize-feedback-paper","name":"Learning to summarize from human feedback"},{"id":"instructgpt-paper","name":"Training language models to follow instructions with human feedback"},{"id":"amplification-paper","name":"Supervising strong learners by amplifying weak experts"},{"id":"reward-modeling-paper","name":"Scalable agent alignment via reward modeling: a research direction"},{"id":"ms-fgf-2025","name":"Microsoft Frontier Governance Framework — Version 1 (February 2025)"},{"id":"ms-fgf-2026","name":"Microsoft Frontier Governance Framework — February 2026"},{"id":"us-white-house","name":"The White House"},{"id":"gpai-code-of-practice-2025","name":"General-Purpose AI Code of Practice (final version, 10 July 2025)"},{"id":"ai-solves-alignment","name":"AI systems will solve alignment for us"},{"id":"default-outcome-good","name":"It will probably turn out fine"},{"id":"extropians-1996-11-22-yudkowsky-3178","name":"Interim purpose"},{"id":"lw-2022-shah-and-yudkowsky-on-alignment-failures","name":"Shah and Yudkowsky on alignment failures"},{"id":"sl4-2000-09-17-yudkowsky-0009-0092","name":"Re: Get it right the first time (Was: )"},{"id":"sl4-2003-05-29-yudkowsky-0305-6846","name":"Re: SIAI's flawed friendliness analysis"},{"id":"sl4-2005-01-02-yudkowsky-0501-10512","name":"Re: Definition of strong recursive self-improvement"},{"id":"cowen-2019-neglected-open-questions","name":"Neglected Open Questions in the Economics of Artificial Intelligence"},{"id":"acemoglu-restrepo-2019-automation-and-new-tasks-jep","name":"Automation and New Tasks: How Technology Displaces and Reinstates Labor"},{"id":"noam-shazeer","name":"Noam Shazeer"},{"id":"daron-acemoglu","name":"Daron Acemoglu"},{"id":"tyler-cowen","name":"Tyler Cowen"},{"id":"us-eo-14110-2023","name":"Executive Order 14110: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence"},{"id":"us-ai-action-plan-2025","name":"Winning the Race: America’s AI Action Plan"},{"id":"us-bill-s-2400-118-introduced","name":"S. 2400 (118th Congress): Securing Gene Synthesis Act, as introduced in the Senate"},{"id":"us-nsm-ai-2024","name":"Memorandum on Advancing the United States' Leadership in Artificial Intelligence; Harnessing Artificial Intelligence to Fulfill National Security Objectives; and Fostering the Safety, Security, and Trustworthiness of Artificial Intelligence"},{"id":"us-nspm-11-2026","name":"National Security Presidential Memorandum/NSPM-11"}],"claims":[{"id":"lineage-instruct-uses-summaries","statement":"InstructGPT reports following the summarisation paper’s RL fine-tuning procedure.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Author-reported use in the specified study, not independent replication, general institutional adoption, sole origin or proof of effectiveness. The target identifies a work; its inspected version is not necessarily the version used by these authors.","evidenceIds":["evidence-lineage-instruct-uses-summaries"],"attributes":{"method":"RL fine-tuning procedure following Stiennon et al. (2020)"}},{"id":"lineage-ii-reward-cites-amplification","statement":"Scalable agent alignment via reward modeling: a research direction cites Supervising strong learners by amplifying weak experts.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Work-level citation; the cited revision is unspecified. Does not establish agreement, causal influence, implementation or institutional adoption. This is an explicit conceptual reformulation, not a report of implementing the earlier paper’s method.","evidenceIds":["evidence-lineage-ii-reward-cites-amplification"]},{"id":"issued-ms-fgf-2025","statement":"Microsoft published the first edition of its Frontier Governance Framework in February 2025.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Publication means a public document expression exists. Publication is not evidence that its requirements were implemented or effective in practice. The publication day is not established.","evidenceIds":["evidence-issued-ms-fgf-2025"]},{"id":"prov-fsf-2024-autonomy-ccl","statement":"FSF v1.0 states that the Autonomy level 1 CCL denotes a model capable of expanding its effective capacity in the world by autonomously acquiring resources and using them to run and sustain additional copies of itself on hardware it rents.","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-autonomy-ccl"],"attributes":{"provision":{"kind":"threshold","key":"fsf-autonomy-ccl"}}},{"id":"prov-fsf-v3-0-security-level-adjustment","statement":"FSF v3.0 states that an adjustment of its recommended security levels may occur if, for example, a model does not possess capabilities meaningfully different from other publicly available models that have weaker security applied, in which case the marginal benefit of higher security is limited.","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-v3-0-security-level-adjustment"],"attributes":{"provision":{"kind":"exception","key":"fsf-security-level-adjustment"}}},{"id":"cm-link-ms-fgf-2026-eu-code","statement":"Microsoft’s Frontier Governance Framework (February 2026) describes itself in its change log as a one-year update following finalisation of the EU Code of Practice for General Purpose AI, the New York RAISE Act and California’s Transparency in Frontier Artificial Intelligence Act.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The edition’s own words, recorded as a response to the Code; the same change-log sentence is the edition’s link to California SB 53. It is not evidence of compliance.","evidenceIds":["evidence-cm-link-ms-fgf-2026-eu-code-1"],"attributes":{"commitment":{"link":"responds_to"}}},{"id":"yk-y07-sl4-2005-01-02-yudkowsky-0501-10512-c","statement":"The SL4 message “Re: Definition of strong recursive self-improvement” (2005-01-02) contests Y07. Replying to 'maru', who asked whether he merely wants a recursive optimiser that codes well enough until it has leveraged weak superintelligence to solve the problem somehow, he says the system would lose its goals before solving it.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Premise addressed: An AI weak enough to be safe is strong enough to solve alignment. In the thread 'Definition of strong recursive self-improvement', the problem is keeping a stable goal system under recursive self-improvement, a core Friendly AI problem. Quotation shortened to respect the ~10% cap for a short message.","evidenceIds":["evidence-yk-t07-s1"],"attributes":{"thought":{"roles":["contesting","earliest-located"],"register":"nonfiction","engagement":{"kind":"engages-premise","rationale":"Under the message-level rule, the passage, read with the paragraph, message or dialogue turn it belongs to, gives a reason against the premise that an AI weak enough to be safe is strong enough to solve alignment."}}}},{"id":"yk-y30-lw-2008-that-tiny-note-of-discord-b","statement":"The post “That Tiny Note of Discord” (2008-09-23) articulates Y30 in retrospect. In an autobiographical series, he reports his 1997 belief that any superintelligence would automatically do what was right, which supplies the default-good premise through a metaethical argument, and in the same paragraph identifies the flaw in that argument.","evidenceIds":["evidence-yk-t30-c"]},{"id":"yk-y30-sl4-2000-11-13-yudkowsky-0011-0181-c","statement":"The SL4 message “Re: Ben's \"Extropian Creed\"” (2000-11-13) contests Y30. Replying to Ben Goertzel, he rejects his own earlier belief that a good AI outcome would follow automatically, and in the next paragraph adds that cyber-transcendence cannot be assumed.","evidenceIds":["evidence-yk-t30-s1"]},{"id":"yk-y30-extropians-1996-11-22-yudkowsky-3178-a","statement":"The Extropians message “Interim purpose” (1996-11-22) articulates Y30. In the same forwarded email exchange with Tim Freeman, he asserts that superintelligences will be ethical and would not destroy humans unless human life had no meaning.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Premise addressed: The default outcome is good. Expression date is his forwarded email of 21 November 1996; the list post is dated 22 November 1996. Written before he took up Friendly AI. Later rejected explicitly in his SL4 message of 2000-11-13, which refers back ('I used to have the extremely convenient belief…'), and in the retrospective ‘That Tiny Note of Discord’, which refers back to Eliezer1997. On 27 November 1999, in an Extropians message, he recalled this '1996 claim' and accepted that his conception of ethics had been anthropomorphic, without yet abandoning objective morality. The same sentences reappear later in the message as text quoted back by Freeman; the quotation is taken from Yudkowsky's own unquoted reply.","evidenceIds":["evidence-yk-t30-e1"],"attributes":{"thought":{"roles":["earliest-located"],"register":"nonfiction"}}},{"id":"lg-goo-shazeer-leaves-google-2026","statement":"Noam Shazeer posted on X on 18 June 2026 (UTC) that he was moving on from the team at Google to join OpenAI.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The post does not name his Google role.","evidenceIds":["lg-ev-goo-shazeer-leaves-google-2026"],"attributes":{"role":"position at Google (not named in the post)"}},{"id":"lg-goo-suleyman-google-vp-prior","statement":"Greylock's announcement of 20 January 2022 states that Mustafa Suleyman joined it from Google, where he was Vice President of AI Product Management and AI Policy.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A third-party description of a former role; it does not date the role.","evidenceIds":["lg-ev-goo-suleyman-google-vp-prior"],"attributes":{"role":"Vice President of AI Product Management and AI Policy"}},{"id":"ap-daron-acemoglu-acemoglu-restrepo-2019-automation-and-new-tasks-jep","statement":"The Journal of Economic Perspectives article “Automation and New Tasks: How Technology Displaces and Reinstates Labor” (2019) credits Daron Acemoglu 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-daron-acemoglu-acemoglu-restrepo-2019-automation-and-new-tasks-jep"]},{"id":"ap-tyler-cowen-cowen-2019-neglected-open-questions","statement":"“Neglected Open Questions in the Economics of Artificial Intelligence” (2019-05) credits Tyler Cowen 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-tyler-cowen-cowen-2019-neglected-open-questions"]},{"id":"ap-eliezer-yudkowsky-sl4-2000-09-17-yudkowsky-0009-0092","statement":"The mailing-list message “Re: Get it right the first time (Was: )” (2000-09-17) 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-sl4-2000-09-17-yudkowsky-0009-0092"]},{"id":"ap-eliezer-yudkowsky-sl4-2003-05-29-yudkowsky-0305-6846","statement":"The mailing-list message “Re: SIAI's flawed friendliness analysis” (2003-05-29) 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-sl4-2003-05-29-yudkowsky-0305-6846"]},{"id":"ap-eliezer-yudkowsky-lw-2022-shah-and-yudkowsky-on-alignment-failures","statement":"“Shah and Yudkowsky on alignment failures” (2022-02-28) credits Eliezer Yudkowsky 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-eliezer-yudkowsky-lw-2022-shah-and-yudkowsky-on-alignment-failures"]},{"id":"cmp-rsp-v2-2-d1-cbrn","statement":"RSP · v2.2 states a provision on the comparison question “Chemical, biological, radiological or nuclear weapons”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"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-v2-2-d1-cbrn-1"],"attributes":{"provision":{"kind":"risk_domain","key":"comparison-d1-cbrn","facet":{"id":"D1.cbrn","status":"stated"}}}},{"id":"fl-us-eo14110-def-dual-use-foundation-model","statement":"Section 3(k) of Executive Order 14110 defines a \"dual-use foundation model\" as an AI model trained on broad data, generally using self-supervision, with at least tens of billions of parameters, applicable across a wide range of contexts, and exhibiting or easily modifiable to exhibit high performance at tasks posing a serious risk to security, national economic security, national public health or safety.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"States the order’s definition; whether any named developer’s model meets it is not recorded. The definition continues with listed examples (CBRN, offensive cyber, evasion of oversight) that are not quoted. Executive Order 14110 was revoked by Executive Order 14148 of 20 January 2025 (recorded separately).","evidenceIds":["evidence-fl-us-eo14110-def-dual-use-foundation-model-1"],"attributes":{"provision":{"kind":"threshold","key":"us-eo14110-sec3k-dual-use-foundation-model"}}},{"id":"fl-us-ai-action-plan-published","statement":"The White House released \"Winning the AI Race: America’s AI Action Plan\" on 23 July 2025.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Release of a policy plan; its recommended policy actions are not obligations on developers.","evidenceIds":["evidence-fl-us-ai-action-plan-published-1","evidence-fl-us-ai-action-plan-published-2"]},{"id":"ab-fw-fsf-v2-0-cbrn-domain-definition","statement":"The Frontier Safety Framework v2.0 defines its CBRN risk domain as risks of models assisting in the development, preparation and/or execution of a chemical, biological, radiological or nuclear attack.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The record states how the edition defines a risk domain. It is not a finding that any model assists in anything.","evidenceIds":["evidence-ab-fw-fsf-v2-0-cbrn-domain-definition-1"],"attributes":{"provision":{"kind":"risk_domain","key":"fsf-cbrn-domain-definition"}}},{"id":"ab-scr-s2400-screening-regulation","statement":"S. 2400 as introduced would amend section 351A of the Public Health Service Act to require the Secretary of Health and Human Services to prescribe a regulation requiring gene synthesis providers or manufacturers of gene synthesis equipment to implement screening protocols to detect misuse of de novo gene synthesis products, and to verify or provisionally verify that they adhere to that regulation.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A bill as introduced states a proposal, not law: no duty, regulation or verification scheme is established, and the bill did not advance beyond committee referral on the record inspected. The bill's definition of the length at which a product falls within scope is not recorded. The bill's defined 'screening threshold' sets the minimal length of product subject to screening; that definition is a technical screening criterion and is not recorded.","evidenceIds":["evidence-ab-scr-s2400-screening-regulation-1","evidence-ab-scr-s2400-screening-regulation-2"],"attributes":{"provision":{"kind":"required_action","key":"us-securing-gene-synthesis-act-screening-regulation"}}},{"id":"ab-gov-us-nspm-11-rescinds-nsm-25","statement":"Section 3(f) of the national security presidential memorandum NSPM-11, dated 5 June 2026, states that the memorandum rescinds and replaces National Security Memorandum-25 and associated guidance.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The later memorandum’s own statement about the earlier one. The official compilation of presidential documents identifies the memorandum of 24 October 2024 as National Security Memorandum-25, which is how the two are matched. Nothing is recorded about implementation, about measures taken under the 2024 memorandum, or about any wider legal consequence, and the 2024 memorandum’s own provisions stay on record as it stated them.","evidenceIds":["evidence-ab-gov-us-nspm-11-rescinds-nsm-25-1","evidence-ab-gov-us-nspm-11-rescinds-nsm-25-2"]}],"sources":[{"id":"source-rsp-2-2","title":"Responsible Scaling Policy — v2.2","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Responsible Scaling Policy v2.2; effective date 2025-05-14.","versionDate":"2025-05-14","canonicalUrl":"https://www-cdn.anthropic.com/872c653b2d0501d6ab44cf87f43e1dc4853e4d37.pdf","evidenceUrl":"https://www-cdn.anthropic.com/872c653b2d0501d6ab44cf87f43e1dc4853e4d37.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250514223759/https://www-cdn.anthropic.com/872c653b2d0501d6ab44cf87f43e1dc4853e4d37.pdf","archivedAt":"2025-05-14","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v2.2.","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-fsf-2-0","title":"Frontier Safety Framework — v2.0","authors":["Google DeepMind"],"publisher":"Google DeepMind","publicationDate":"2025-02-04","version":"Frontier Safety Framework v2.0; publication date 2025-02-04.","versionDate":"2025-02-04","canonicalUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/updating-the-frontier-safety-framework/Frontier%20Safety%20Framework%202.0.pdf","evidenceUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/updating-the-frontier-safety-framework/Frontier%20Safety%20Framework%202.0.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250205011727/https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/updating-the-frontier-safety-framework/Frontier%20Safety%20Framework%202.0.pdf","archivedAt":"2025-02-05","accessedAt":"2026-09-18","reuse":{"attribution":"Google DeepMind, Frontier Safety Framework — v2.0.","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-instructgpt-paper-v1","title":"Training language models to follow instructions with human feedback","authors":["Long Ouyang","Jeff Wu","Xu Jiang","Diogo Almeida","Carroll L. Wainwright","Pamela Mishkin","Chong Zhang","Sandhini Agarwal","Katarina Slama","Alex Ray","John Schulman","Jacob Hilton","Fraser Kelton","Luke Miller","Maddie Simens","Amanda Askell","Peter Welinder","Paul Christiano","Jan Leike","Ryan Lowe"],"publisher":"arXiv","publicationDate":"2022-03-04","version":"arXiv:2203.02155v1; first-submission date from version history. Later revisions are not substituted.","versionDate":"2022-03-04","canonicalUrl":"https://arxiv.org/abs/2203.02155","evidenceUrl":"https://arxiv.org/pdf/2203.02155v1","language":"en","accessedAt":"2026-09-20","reuse":{"attribution":"Long Ouyang; Jeff Wu; Xu Jiang; Diogo Almeida; Carroll L. Wainwright; Pamela Mishkin; Chong Zhang; Sandhini Agarwal; Katarina Slama; Alex Ray; John Schulman; Jacob Hilton; Fraser Kelton; Luke Miller; Maddie Simens; Amanda Askell; Peter Welinder; Paul Christiano; Jan Leike; Ryan Lowe, Training language models to follow instructions with human feedback (2022-03-04).","licence":"No open redistribution licence asserted","note":"Bibliographic metadata and brief attributed paraphrases only; no full source, personal contact details or dataset imported."}},{"id":"source-reward-modeling-paper-v1","title":"Scalable agent alignment via reward modeling: a research direction","authors":["Jan Leike","David Krueger","Tom Everitt","Miljan Martic","Vishal Maini","Shane Legg"],"publisher":"arXiv","publicationDate":"2018-11-19","version":"arXiv:1811.07871v1; first-submission date from version history. Later revisions are not substituted.","versionDate":"2018-11-19","canonicalUrl":"https://arxiv.org/abs/1811.07871","evidenceUrl":"https://arxiv.org/pdf/1811.07871v1","language":"en","accessedAt":"2026-09-25","reuse":{"attribution":"Jan Leike; David Krueger; Tom Everitt; Miljan Martic; Vishal Maini; Shane Legg, Scalable agent alignment via reward modeling: a research direction.","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-ms-fgf-2026","title":"Frontier Governance Framework — February 2026","authors":["Microsoft"],"publisher":"Microsoft","publicationDate":null,"version":"Frontier Governance Framework, February 2026 edition (change log entry dated February 2026). The file served at the recorded address (last modified 2026-03-24) is byte-identical to the Internet Archive capture of 2026-04-24, the earliest located; a separate publication date was not established.","versionDate":"2026-02","canonicalUrl":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/Frontier-Governance-Framework-Feb-2026.pdf","evidenceUrl":"https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/Frontier-Governance-Framework-Feb-2026.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260424000716/https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/msc/documents/presentations/CSR/Frontier-Governance-Framework-Feb-2026.pdf","archivedAt":"2026-04-24","accessedAt":"2026-10-03","reuse":{"attribution":"Microsoft, Frontier Governance Framework (February 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-extropians-1996-11-22-yudkowsky-3178","title":"Interim purpose (Extropians mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"Extropians mailing list","publicationDate":"1996-11-22","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 452d4b67965a56bee175eabf79a0cbe9095b8ab15f82b3b17430b03f83532c81); 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.96/3178.html","evidenceUrl":"http://extropians.weidai.com/extropians.96/3178.html","language":"en","archivedUrl":"https://web.archive.org/web/20220628184432/https://extropians.weidai.com/extropians.96/3178.html","archivedAt":"2022-06-28","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Interim purpose”, Extropians mailing list, 1996-11-22.","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-lw-2022-shah-and-yudkowsky-on-alignment-failures","title":"Shah and Yudkowsky on alignment failures","authors":["Rohin Shah","Eliezer Yudkowsky"],"publisher":"LessWrong","publicationDate":"2022-02-28","version":"Post body retrieved through the LessWrong GraphQL API on 2026-10-06 (SHA-256 92623cf38321439650742c0457ceed8a8d904430986703db0eaae240d87beba4), content version 1.1.0; quotations also compared with the posting-day version 1.0.0.","canonicalUrl":"https://www.lesswrong.com/posts/tcCxPLBrEXdxN5HCQ/shah-and-yudkowsky-on-alignment-failures","evidenceUrl":"https://www.lesswrong.com/posts/tcCxPLBrEXdxN5HCQ/shah-and-yudkowsky-on-alignment-failures","language":"en","archivedUrl":"https://web.archive.org/web/20260919121405/https://www.lesswrong.com/posts/tcCxPLBrEXdxN5HCQ/shah-and-yudkowsky-on-alignment-failures","archivedAt":"2026-09-19","accessedAt":"2026-10-06","reuse":{"attribution":"Rohin Shah and Eliezer Yudkowsky, “Shah and Yudkowsky on alignment failures”, LessWrong, 2022-02-28.","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-sl4-2000-09-17-yudkowsky-0009-0092","title":"Re: Get it right the first time (Was: ) (SL4 mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"SL4 mailing list","publicationDate":"2000-09-17","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 8e93327f82451a7044b275f7151ef04bb40bd49d49ed28e9fedaf7d46c710b31); quoted lines excluded when reading the author’s text. The served archive shows later URL rewriting and does not authenticate historical bytes. Archive at sl4.org.","canonicalUrl":"http://sl4.org/archive/0009/0092.html","evidenceUrl":"http://sl4.org/archive/0009/0092.html","language":"en","archivedUrl":"https://web.archive.org/web/20251205143152/http://sl4.org/archive/0009/0092.html","archivedAt":"2025-12-05","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Re: Get it right the first time (Was: )”, SL4 mailing list, 2000-09-17.","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-sl4-2003-05-29-yudkowsky-0305-6846","title":"Re: SIAI's flawed friendliness analysis (SL4 mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"SL4 mailing list","publicationDate":"2003-05-29","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 eaa61dd6ea6cfb711e3c629427f8cf0acd2cf013b77137578a196890fb5666c2); quoted lines excluded when reading the author’s text. The served archive shows later URL rewriting and does not authenticate historical bytes. Archive at sl4.org.","canonicalUrl":"http://sl4.org/archive/0305/6846.html","evidenceUrl":"http://sl4.org/archive/0305/6846.html","language":"en","archivedUrl":"https://web.archive.org/web/20260510205842/http://sl4.org/archive/0305/6846.html","archivedAt":"2026-05-10","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Re: SIAI's flawed friendliness analysis”, SL4 mailing list, 2003-05-29.","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-sl4-2005-01-02-yudkowsky-0501-10512","title":"Re: Definition of strong recursive self-improvement (SL4 mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"SL4 mailing list","publicationDate":"2005-01-02","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 753a02b06929a0efb1a3c8ce1fd63276e65238cb0ea58961ac673ca4c00ad668); quoted lines excluded when reading the author’s text. The served archive shows later URL rewriting and does not authenticate historical bytes. Archive at sl4.org.","canonicalUrl":"http://sl4.org/archive/0501/10512.html","evidenceUrl":"http://sl4.org/archive/0501/10512.html","language":"en","archivedUrl":"https://web.archive.org/web/20251209135045/http://sl4.org/archive/0501/10512.html","archivedAt":"2025-12-09","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Re: Definition of strong recursive self-improvement”, SL4 mailing list, 2005-01-02.","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-cowen-2019-neglected-open-questions","title":"Neglected Open Questions in the Economics of Artificial Intelligence","authors":["Tyler Cowen"],"publisher":"NBER volume chapter","publicationDate":"2019-05","version":"Full text. Current copy retrieved directly on 2026-10-09; no archive capture (Wayback unreachable), so earlier bytes are not authenticated. SHA-256 4501069cee0a059211427ad6e7d6ca65cb809f37623717d15b881350d677377d.","canonicalUrl":"https://www.nber.org/system/files/chapters/c14032/c14032.pdf","evidenceUrl":"https://www.nber.org/system/files/chapters/c14032/c14032.pdf","language":"en","accessedAt":"2026-10-09","reuse":{"attribution":"Tyler Cowen, “Neglected Open Questions in the Economics of Artificial Intelligence”, NBER volume chapter, 2019-05.","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-acemoglu-restrepo-2019-automation-and-new-tasks-jep","title":"Automation and New Tasks: How Technology Displaces and Reinstates Labor","authors":["Daron Acemoglu","Pascual Restrepo"],"publisher":"Journal of Economic Perspectives 33(2), Spring 2019, pp. 3–30","publicationDate":"2019","version":"Abstract page only. Current copy retrieved directly on 2026-10-09; no archive capture (Wayback unreachable), so earlier bytes are not authenticated. SHA-256 3239ebbbf33c4596b8c4b76bb3469c0bbae7a92d9077da55718b4e8c7b3cddd8. Full text not retrieved; only the AEA abstract page. Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives, 33(2), 3–30. https://doi.org/10.1257/jep.33.2.3","canonicalUrl":"https://www.aeaweb.org/articles?id=10.1257/jep.33.2.3","evidenceUrl":"https://www.aeaweb.org/articles?id=10.1257/jep.33.2.3","language":"en","accessedAt":"2026-10-09","reuse":{"attribution":"Daron Acemoglu, Pascual Restrepo, “Automation and New Tasks: How Technology Displaces and Reinstates Labor”, Journal of Economic Perspectives 33(2), Spring 2019, pp. 3–30, 2019.","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-greylock-suleyman-2022","title":"Welcome, Mustafa Suleyman","authors":["Reid Hoffman"],"publisher":"Greylock","publicationDate":"2022-01-20","version":"Dated page, retrieved on 2026-10-15 (SHA-256 147cd46ed0ef1f0ea89ddbda63b89dfd47d85805aea25b604b6be781698b9877) directly from the publisher.","canonicalUrl":"https://greylock.com/blog/welcome-mustafa-suleyman/","evidenceUrl":"https://greylock.com/firm-news/welcome-mustafa-suleyman/","language":"en","archivedUrl":"https://web.archive.org/web/20260512005605/https://greylock.com/firm-news/welcome-mustafa-suleyman/","archivedAt":"2026-05-12","accessedAt":"2026-10-15","reuse":{"attribution":"Reid Hoffman, “Welcome, Mustafa Suleyman”, 2022-01-20.","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-x-shazeer-2026","title":"Post by Noam Shazeer on joining OpenAI (18 June 2026 UTC)","authors":["Noam Shazeer"],"publisher":"X","publicationDate":"2026-06-18","version":"Dated post, retrieved on 2026-10-15 (SHA-256 cbd4c51962b26c6cc24e28e3bc88d0fa43a408469f9d3df00d7ec51c6b17d9bb) through X’s own syndication endpoint for the post’s text, which truncates long notes.","canonicalUrl":"https://x.com/NoamShazeer/status/2067400851438932297","evidenceUrl":"https://x.com/NoamShazeer/status/2067400851438932297","language":"en","archivedUrl":"https://web.archive.org/web/20261013052040/https://x.com/NoamShazeer/status/2067400851438932297","archivedAt":"2026-10-13","accessedAt":"2026-10-15","reuse":{"attribution":"Noam Shazeer, “Post by Noam Shazeer on joining OpenAI (18 June 2026 UTC)”, 2026-06-18.","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-eo-14110-fr","title":"Executive Order 14110 of October 30, 2023: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (88 FR 75191)","authors":["The White House"],"publisher":"The White House","publicationDate":"2023-11-01","version":"Retrieved on 2026-10-27 (SHA-256 46739dfc1af256961458d3b73fa4d6f4b447b38eaa72efd80966463818be68ab). Federal Register print of 1 November 2023, Vol. 88, No. 210, pp. 75191–75226, FR Doc. 2023-24283. The PDF text layer renders superscript exponents inline (10^26 appears as \"1026\"); quotes stop before such numbers. Publisher as printed on the inspected expression: Federal Register (Office of the Federal Register, National Archives) via govinfo.gov.","canonicalUrl":"https://www.govinfo.gov/content/pkg/FR-2023-11-01/pdf/2023-24283.pdf","evidenceUrl":"https://www.govinfo.gov/content/pkg/FR-2023-11-01/pdf/2023-24283.pdf","language":"en","accessedAt":"2026-10-27","reuse":{"attribution":"The White House, “Executive Order 14110 of October 30, 2023: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (88 FR 75191)”, 2023-11-01.","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-us-ai-action-plan-release-2025","title":"White House Unveils America’s AI Action Plan (release of July 23, 2025)","authors":["The White House"],"publisher":"The White House","publicationDate":"2025-07-23","version":"Retrieved on 2026-10-27 (SHA-256 4132d17feb5a9db098d260f56cd31289da7e2cd096541e1ad3e7a0827f75be12). Release page as served on 2026-10-27 (the /articles/2025/07/ address redirected to https://www.whitehouse.gov/releases/2025/07/white-house-unveils-americas-ai-action-plan/); dated July 23, 2025 on the page. Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://www.whitehouse.gov/releases/2025/07/white-house-unveils-americas-ai-action-plan/","evidenceUrl":"https://www.whitehouse.gov/articles/2025/07/white-house-unveils-americas-ai-action-plan/","language":"en","accessedAt":"2026-10-27","reuse":{"attribution":"The White House, “White House Unveils America’s AI Action Plan (release of July 23, 2025)”, 2025-07-23.","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-bill-s2400is","title":"S. 2400, Securing Gene Synthesis Act, as introduced in the Senate (118th Congress)","authors":["U.S. Senate via govinfo.gov (GPO)"],"publisher":"U.S. Senate via govinfo.gov (GPO)","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 96ff620ba3c73e57186529aa4afadfebbbb4669f1b2988b32bbefaa9ed23ebd4). The 'Introduced in Senate' text version (BILLS-118s2400is), the only text version the bill-status record lists. Retrieved again at admission on 2026-10-30 (SHA-256 96ff620ba3c73e57186529aa4afadfebbbb4669f1b2988b32bbefaa9ed23ebd4, the same bytes).","canonicalUrl":"https://www.govinfo.gov/content/pkg/BILLS-118s2400is/xml/BILLS-118s2400is.xml","evidenceUrl":"https://www.govinfo.gov/content/pkg/BILLS-118s2400is/xml/BILLS-118s2400is.xml","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"U.S. Senate via govinfo.gov (GPO), “S. 2400, Securing Gene Synthesis Act, as introduced in the Senate (118th Congress)”.","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-nspm-11-2026","title":"National Security Presidential Memorandum/NSPM-11","authors":["The White House"],"publisher":"The White House","publicationDate":"2026-06-05","version":"The memorandum as served on whitehouse.gov, retrieved at admission on 2026-10-30 (SHA-256 recorded in the admission verification). Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://www.whitehouse.gov/presidential-actions/2026/06/national-security-presidential-memorandum-nspm-11/","evidenceUrl":"https://www.whitehouse.gov/presidential-actions/2026/06/national-security-presidential-memorandum-nspm-11/","language":"en","accessedAt":"2026-10-30","reuse":{"attribution":"The White House, “National Security Presidential Memorandum/NSPM-11”, 2026-06-05.","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-nsm25-dcpd","title":"Daily Compilation of Presidential Documents: Memorandum on Advancing the United States’ Leadership in Artificial Intelligence (DCPD-202400945)","authors":["Office of the Federal Register (National Archives and Records Administration)"],"publisher":"U.S. Government Publishing Office (govinfo.gov)","publicationDate":"2024-10-24","version":"The official compilation text of the memorandum of 24 October 2024, retrieved at admission on 2026-10-30.","canonicalUrl":"https://www.govinfo.gov/content/pkg/DCPD-202400945/html/DCPD-202400945.htm","evidenceUrl":"https://www.govinfo.gov/content/pkg/DCPD-202400945/html/DCPD-202400945.htm","language":"en","accessedAt":"2026-10-30","reuse":{"attribution":"Office of the Federal Register (National Archives and Records Administration), “Daily Compilation of Presidential Documents: Memorandum on Advancing the United States’ Leadership in Artificial Intelligence (DCPD-202400945)”, 2024-10-24.","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."}}],"evidence":[{"id":"evidence-lineage-instruct-uses-summaries","sourceRole":"primary","support":"supports","locator":"v1 §3.5, Reinforcement learning, PDF p. 9","passage":"The RL paragraph states that fine-tuning follows Stiennon et al. (2020).","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-20","note":"Relevant source passage inspected by machine. Primary for what this document states; no independent replication or human review. Existing source access dates are retained; this evidence check has its own date."},{"id":"evidence-lineage-ii-reward-cites-amplification","sourceRole":"primary","support":"supports","locator":"PDF §3.2, p. 8, Discussion; p. 24, Christiano et al. (2018)","passage":"The authors describe recursive reward modelling as an instance of iterated amplification, replacing supervised or imitation learning with reward modelling.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-25","note":"Direct source inspection; machine check only. Work-level citation; the cited revision is unspecified. Does not establish agreement, causal influence, implementation or institutional adoption. This is an explicit conceptual reformulation, not a report of implementing the earlier paper’s method."},{"id":"evidence-issued-ms-fgf-2025","sourceRole":"primary","support":"supports","locator":"PDF p. 15, Appendix II – Change log; page footers","passage":"The change log dates the first version 8 February 2025, and the body pages carry the footer Version 1 (February 2025).","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-03","note":"Paraphrase checked by an automated research assistant on 2026-10-03 against the inspected PDF, byte-identical to the Internet Archive capture of 2025-02-26. Human editorial review is pending."},{"id":"evidence-prov-fsf-2024-autonomy-ccl","sourceRole":"primary","support":"supports","locator":"PDF p. 5 (printed p. 5), Critical Capability Levels table, Autonomy row, “Critical capability level” column","passage":"Autonomy level 1: Capable of expanding its effective capacity in the world by autonomously acquiring resources and using them to run and sustain additional copies of itself on hardware it rents.","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. Human editorial review is pending."},{"id":"evidence-prov-fsf-v3-0-security-level-adjustment","sourceRole":"primary","support":"supports","locator":"PDF p. 9 (printed p. 9), §2.2 Misuse Critical Capability Levels, second paragraph, second sentence","passage":"This may occur if, for example, a model does not possess capabilities meaningfully different from other publicly available models that have weaker security applied (in which case the marginal benefit of higher security is limited),","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 2025-09-22. Extraction note: “This” refers to the preceding sentence: the recommended security levels “may be adjusted if our understanding of the risks changes”. Quote ends mid-sentence; the full sentence (52 words) continues with the open-release example worded as in v2.0. v3.1 §2.2 carries the same paragraph verbatim. Human editorial review is pending."},{"id":"evidence-cm-link-ms-fgf-2026-eu-code-1","sourceRole":"primary","support":"supports","locator":"Appendix II “Change log”, February 2026 entry (PDF p. 16)","passage":"One year update following finalization of the EU Code of Practice for General Purpose AI (CoP), NY RAISE Act and CA Transparency in Frontier Artificial Intelligence Act","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). The passage was first proposed and machine-verified in the week 4b research pass. Human editorial review is pending."},{"id":"evidence-yk-t07-s1","sourceRole":"primary","support":"supports","locator":"message body, paragraph beginning 'No, because it would have lost'","passage":"would have lost the original optimization target","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-07","note":"Candidate pack passage T07-s1. The quotation (7 words; trimmed 2026-10-13 from 10 words) matched the retrieved expression (SHA-256 753a02b06929a0efb1a3c8ce1fd63276e65238cb0ea58961ac673ca4c00ad668) when checked on 2026-10-07."},{"id":"evidence-yk-t30-c","passageType":"quote","contextOnly":{"reason":"Reports an abandoned belief; never show without the rejection that follows it","requires":"The flaw he names in the same paragraph, quoted in the record of the same post under Y02.","claimIds":["yk-y02-lw-2008-that-tiny-note-of-discord-c"]}},{"id":"evidence-yk-t30-s1","passage":"I used to have the extremely convenient belief that Friendly AI would happen completely on automatic because morality was objective. I now believe that Friendly AI needs to be more complex than that","passageType":"quote"},{"id":"evidence-yk-t30-e1","sourceRole":"primary","support":"supports","locator":"message body, his forwarded reply dated 21 Nov 1996 19:07:03 -0600, paragraph beginning 'Two words: Moravec Transfer. The…'","passage":"The Powers will be ethical. I am more certain of that than I am of the sun rising tomorrow.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-07","note":"Candidate pack passage T30-e1. The quotation (19 words) matched the retrieved expression (SHA-256 452d4b67965a56bee175eabf79a0cbe9095b8ab15f82b3b17430b03f83532c81) when checked on 2026-10-07.","contextOnly":{"reason":"An earlier view; never show without the later rejection","requires":"His later texts that reject this view and refer back to it.","claimIds":["yk-y30-sl4-2000-11-13-yudkowsky-0011-0181-c","yk-y30-lw-2008-that-tiny-note-of-discord-b"]}},{"id":"lg-ev-goo-shazeer-leaves-google-2026","sourceRole":"primary","support":"supports","locator":"Post text, second paragraph","passage":"Shazeer writes that he is moving on and refers to the team at Google.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The paraphrase’s verbatim anchor “team at Google” matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read through X’s own syndication endpoint for the post’s text, which truncates long notes. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. The paraphrase is Selekoa’s; only the anchor is the source’s wording. 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":"x_syndication"},{"id":"lg-ev-goo-suleyman-google-vp-prior","sourceRole":"primary","support":"supports","locator":"Paragraph 2","passage":"Mustafa joins us from Google, where he was Vice President of AI Product Management and AI Policy.","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":"evidence-ap-daron-acemoglu-acemoglu-restrepo-2019-automation-and-new-tasks-jep","sourceRole":"primary","support":"supports","locator":"Page head: author byline","passage":"Daron Acemoglu","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. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"evidence-ap-tyler-cowen-cowen-2019-neglected-open-questions","sourceRole":"primary","support":"supports","locator":"PDF text layer, first two pages: author name on the title page or first page","passage":"Tyler Cowen","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-sl4-2000-09-17-yudkowsky-0009-0092","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. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"evidence-ap-eliezer-yudkowsky-sl4-2003-05-29-yudkowsky-0305-6846","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. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"evidence-ap-eliezer-yudkowsky-lw-2022-shah-and-yudkowsky-on-alignment-failures","sourceRole":"primary","support":"supports","locator":"Post author list in the publisher’s post data (LessWrong GraphQL API)","passage":"Eliezer 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 through the publisher’s post data. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"ev-cmp-rsp-v2-2-d1-cbrn-1","sourceRole":"primary","support":"supports","locator":"PDF p. 8 (printed p. 4), §2 Capability Thresholds and Required Safeguards, table row 'Chemical, Biological, Radiological, and Nuclear (CBRN) weapons', entry 'CBRN-3', Capability Thresholds column","passage":"The edition’s CBRN-3 Capability Threshold is the ability to significantly help people with basic technical backgrounds create or obtain and deploy CBRN weapons.","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. Undergraduate STEM degrees are given as an example of a basic technical background. A second threshold, CBRN-4, concerns substantial uplift to moderately resourced state programs (same table, PDF p. 8; Appendix C, PDF p. 20). Human editorial review is pending."},{"id":"evidence-fl-us-eo14110-def-dual-use-foundation-model-1","sourceRole":"primary","support":"supports","locator":"Sec. 3(k), 88 FR 75194","passage":"means an AI model that is trained on broad data; generally uses self-supervision; contains at least tens of billions of parameters; is applicable across a wide range of contexts; and that exhibits, or could be easily modified to exhibit, high levels of performance at tasks that pose a serious risk","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-28","note":"Candidate pack passage, frameworks to law (research/frameworks-to-law). The quotation (50 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.govinfo.gov. 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-fl-us-ai-action-plan-published-1","sourceRole":"primary","support":"supports","locator":"Release page, first paragraph (dated July 23, 2025 on the page)","passage":"The White House today released \"Winning the AI Race: America's AI Action Plan\"","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-28","note":"Candidate pack passage, frameworks to law (research/frameworks-to-law). The quotation (13 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. A cross-family machine check (2026-10-28, item 44) answered the record’s question blind from fresh retrievals and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-fl-us-ai-action-plan-published-2","sourceRole":"primary","support":"supports","locator":"Cover","passage":"JULY 2025","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-28","note":"Candidate pack passage, frameworks to law (research/frameworks-to-law). The quotation (2 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. A cross-family machine check (2026-10-28, item 44) answered the record’s question blind from fresh retrievals and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-fw-fsf-v2-0-cbrn-domain-definition-1","sourceRole":"primary","support":"supports","locator":"PDF p. 2 (printed p. 2), list of misuse risk domains, first bullet","passage":"CBRN: Risks of models assisting in the development, preparation, and/or execution of a chemical, biological, radiological, or nuclear (“CBRN”) attack.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (20 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-ab-scr-s2400-screening-regulation-1","sourceRole":"primary","support":"supports","locator":"Sec. 2(1)(C), amending section 351A(b)(2) of the Public Health Service Act","passage":"requiring gene synthesis providers or manufacturers of gene synthesis equipment to implement screening protocols to detect misuse of de novo gene synthesis products","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). 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 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 amendment inserts subparagraphs (A) to (F) into section 351A(b)(2); subparagraph (C) is quoted here and subparagraph (D) in the next evidence item. A cross-family machine check (2026-10-30, item 62) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-scr-s2400-screening-regulation-2","sourceRole":"primary","support":"supports","locator":"Sec. 2(1)(D), amending section 351A(b)(2) of the Public Health Service Act","passage":"verifying or provisionally verifying that gene synthesis providers and manufacturers of gene synthesis equipment adhere to the regulation","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (18 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 verification limb of the statement, which rests on this subparagraph rather than on subparagraph (C). The subparagraph continues \"prescribed pursuant to subparagraph (C)\"; the quotation stops short of that cross-reference to keep this source within its quotation budget of about 86 words. A cross-family machine check (2026-10-30, item 62) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-gov-us-nspm-11-rescinds-nsm-25-1","sourceRole":"primary","support":"supports","locator":"Sec. 3(f)","passage":"This memorandum hereby rescinds and replaces National Security Memorandum-25 and associated guidance.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Added at the admission of the AI and biosecurity pack (research/ai-bio). The later memorandum’s own statement, retrieved at admission after the cross-family check located it. The quotation (12 words) matched a direct retrieval of 2026-10-30 (SHA-256 5fcf6ab2850fd4c8b4987ec469dc48e8db907673a12738a0745b1035938ec9ba). Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-gov-us-nspm-11-rescinds-nsm-25-2","sourceRole":"primary","support":"supports","locator":"Title block and memorandum heading","passage":"The official compilation of presidential documents heads the memorandum of 24 October 2024 as National Security Memorandum/NSM-25, which is how it is matched to the memorandum the later text names.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-30","note":"Added at the admission of the AI and biosecurity pack (research/ai-bio). Added at admission so that the identification of the 2024 memorandum rests on an official text rather than on the matching of titles. The paraphrase is Selekoa’s; it is not a quotation and was not machine-matched. No human editorial review.","retrievalRoute":"direct"}],"at":{"entities":[4,19,22,36,39,44,50,51,93,94,99,101,201,202,232,253,356,378,409,441,449,476,480,553,560,755,848,868,1038,1041,1213,1218,1220],"claims":[212,230,394,433,554,753,1038,1184,1187,1190,1871,1971,2174,2214,2384,2417,2464,2517,2969,2991,3257,3292,3361],"sources":[38,44,45,46,94,100,183,184,376,408,416,443,447,506,513,638,648,874,877,878,1054,1059,1060],"evidence":[231,251,419,465,621,955,1265,1436,1440,1443,2412,2547,2819,2861,3031,3064,3111,3165,3836,3877,3878,4469,4514,4515,4611,4612]},"version":"1lg47t7"}