{"entities":[{"id":"eliezer-yudkowsky","name":"Eliezer Yudkowsky"},{"id":"risks-v3","name":"Risks from Learned Optimization (v3)"},{"id":"deceptive-alignment","name":"Deceptive alignment"},{"id":"rsp-v3-0","name":"Anthropic Responsible Scaling Policy — v3.0"},{"id":"rsp-v3-1","name":"Anthropic Responsible Scaling Policy — v3.1"},{"id":"rsp-v3-4","name":"Anthropic Responsible Scaling Policy — v3.4"},{"id":"daniel-gross","name":"Daniel Gross"},{"id":"ssi","name":"Safe Superintelligence Inc."},{"id":"ms-fgf-2025","name":"Microsoft Frontier Governance Framework — Version 1 (February 2025)"},{"id":"openai-inc","name":"OpenAI, Inc. (OpenAI Foundation)"},{"id":"g42","name":"G42"},{"id":"frontier-ai-safety-commitments-2024","name":"Frontier AI Safety Commitments, AI Seoul Summit 2024"},{"id":"michie-1968-computer-servant-or-master","name":"Computer—servant or master"},{"id":"levels-of-organization-in-general-intelligence","name":"Levels of Organization in General Intelligence"},{"id":"seed-ai","name":"Seed AI"},{"id":"control-or-purpose-specification","name":"Control or purpose specification"},{"id":"lw-2025-re-recent-anthropic-safety-research","name":"Re: recent Anthropic safety research"},{"id":"sl4-2005-07-13-yudkowsky-0507-11477","name":"Re: Fighting UFAI"},{"id":"eu-gpai-guidelines-2025","name":"Commission Guidelines on the scope of the obligations for providers of general-purpose AI models established by Regulation (EU) 2024/1689 (C(2025) 7719 final)"},{"id":"cyberspace-administration-of-china","name":"Cyberspace Administration of China (国家互联网信息办公室)"},{"id":"ai-productivity-over-next-decade","name":"AI’s productivity effect over the next decade"},{"id":"aghion-jones-jones-2017-ai-and-economic-growth","name":"Artificial Intelligence and Economic Growth"},{"id":"leontief-duchin-1984-impacts-of-automation","name":"The Impacts of Automation on Employment, 1963–2000. Final Report"},{"id":"baidu","name":"Baidu / 百度"},{"id":"character-ai","name":"Character.AI"},{"id":"moonshot-ai","name":"Moonshot AI / 月之暗面"},{"id":"noam-shazeer","name":"Noam Shazeer"},{"id":"shivon-zilis","name":"Shivon Zilis"},{"id":"stepfun","name":"StepFun / 阶跃星辰"},{"id":"zhang-xiangyu","name":"Zhang Xiangyu"},{"id":"charles-i-jones","name":"Charles I. Jones"},{"id":"ernie-3-titan-report","name":"ERNIE 3.0 Titan: Exploring Larger-scale Knowledge Enhanced Pre-training for Language Understanding and Generation"},{"id":"kimi-k15-readme","name":"Kimi-k1.5 repository README"},{"id":"kimi-k2-licence","name":"Kimi K2 Modified MIT License"},{"id":"kimi-k3-licence","name":"Kimi K3 License"},{"id":"cn-cac-genai-filings-announcement-2025","name":"CAC announcement publishing 2024 information on filed generative AI services (国家互联网信息办公室关于发布2024年生成式人工智能服务已备案信息的公告), January 2025"},{"id":"claude-sonnet-4-5-system-card","name":"System Card: Claude Sonnet 4.5"},{"id":"llama-3-herd-paper","name":"The Llama 3 Herd of Models (arXiv:2407.21783)"}],"claims":[{"id":"risks-discusses-deception-2021","statement":"Version 3 of Risks from Learned Optimization discusses deceptive alignment in section 4.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The paper was first submitted in 2019, but this claim is scoped to the inspected 2021 HTML version. No earliest-use claim is made.","evidenceIds":["evidence-risks-discusses-deception-2021"]},{"id":"supersedes-rsp-3-1","statement":"Responsible Scaling Policy v3.1 replaces v3.0 in the publisher’s documented framework version sequence.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"This records replacement in the public policy version sequence, supported by the publisher’s update history. It does not retire historical claims or prove implementation.","evidenceIds":["evidence-supersedes-rsp-3-1","evidence-supersedes-rsp-3-1-2"]},{"id":"gross-ssi-departure-announced-2025","statement":"SSI announced Daniel Gross’s departure on July 3, 2025, stating that he was no longer part of SSI as of June 29.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The claim’s event is the announcement, not June 29. The passage does not specify his departed job title or termination of all legal/equity ties.","evidenceIds":["evidence-gross-ssi-departure-announced-2025"],"attributes":{"role":"Company team member"}},{"id":"cm-seoul-g42","statement":"G42 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-g42-1"],"attributes":{"commitment":{"stage":"listed"}}},{"id":"ts-b6-michie-1968-contests","statement":"Michie’s essay divides tasks among human value decisions, machine detail work and cooperation, and expects computers to work as assistants rather than simply as masters or servants.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A contrary forecast, not a demonstration that advanced machines stay controllable.","evidenceIds":["evidence-ts-b6-michie-1968-contests"],"attributes":{"thought":{"roles":["contesting"],"register":"nonfiction","qualifier":"Disputes the servant-or-master framing.","attested":{"date":"1986","note":"Wording checked in the reprint, chapter 20 of On Machine Intelligence, 2nd edition (1986). The author says little was revised; the 1968 text was not compared."},"engagement":{"kind":"contrary-only","rationale":"States the contrary forecast that computers will be helpers in a division of labour in which people keep decisions of value, without giving a reason why people could not lose the power to direct or stop them."}}}},{"id":"ts-a2-logi","statement":"Levels of Organization in General Intelligence defines a seed AI as an AI designed for self-understanding, self-modification and recursive self-improvement, and treats recursive self-improvement as a consequence of seed AI rather than a cheap way to achieve AI.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The fullest seed-AI treatment located: it distinguishes weak from strong self-improvement and describes \"epochs\" of recursive self-improvement, cautioning that early seed AI is dominated by its human programmers. An editorial judgement among the located texts.","evidenceIds":["evidence-ts-a2-logi"],"attributes":{"thought":{"roles":["most-developed"],"register":"nonfiction","attested":{"note":"MIRI reprint marked \"This version contains minor changes\"; not compared with the 2007 chapter."}}}},{"id":"eugl-non-binding","statement":"The guidelines state that they are not binding for providers of general-purpose AI models, that an authoritative interpretation of the AI Act may only be given by the Court of Justice of the European Union, and that they set out the Commission’s interpretation and application of the AI Act, on which it will base its enforcement action.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A statement about the status of the guidelines, in the guidelines themselves. It states what the text provides, not whether any named developer falls within it or complies. A document’s statement is not evidence of implementation, compliance or enforcement.","evidenceIds":["evidence-eugl-non-binding-1","evidence-eugl-non-binding-2"],"attributes":{"provision":{"kind":"decision_authority","key":"eugl-status","parties":[{"role":"decides","actor":"Court of Justice of the European Union (authoritative interpretation of the AI Act)"}]}}},{"id":"ec-m08-leontief-duchin-1984-impacts-of-automation-b","statement":"Leontief and Duchin, “The Impacts of Automation on Employment, 1963–2000. Final Report” (1984-04) bears on M08. Speaker: Wassily Leontief and Faye Duchin (economists). Headline finding of the report.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Quantitative claim; recheck figures against the printed report before use. Matches the cached .txt.","evidenceIds":["evidence-ec-m08-01"],"attributes":{"thought":{"roles":["context"],"register":"nonfiction","missing":"A labour-saving estimate over 20 years (to 2000), not a productivity or GDP estimate over about ten years, and it concerns computer-based automation, not AI.","tier":{"value":1,"basis":"Report by economists (Leontief and Duchin, NYU Institute for Economic Analysis)."},"conditions":"Model-based projection (dynamic input-output model, 89 sectors, 53 occupations, scenario S3). The abstract adds consumption growth of about 2% a year through the 1980s and 0.5-1.1% through the 1990s, assuming full use of the projected labour force."}}},{"id":"lg-oai-shivon-zilis-nonprofit-board-2019","statement":"OpenAI’s OpenAI LP announcement of 11 March 2019 lists Shivon Zilis as a non-employee member of the OpenAI Nonprofit board.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A board list as of 11 March 2019 (the source’s footnote); it does not establish when the person joined or left the board. The source gives the nonprofit’s legal name as OpenAI Inc.","evidenceIds":["lg-ev-oai-shivon-zilis-nonprofit-board-2019","lg-ev-oai-shivon-zilis-nonprofit-board-2019-inc"],"attributes":{"role":"Board member (non-employee)"}},{"id":"lg-deal-shazeer-character-ai-ceo-2024","statement":"TechCrunch reported on 2 August 2024 that Noam Shazeer was co-founder and CEO of Character.AI.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The record rests on reporting (TechCrunch); no Character.AI primary document stating Shazeer's CEO title was retrieved.","evidenceIds":["lg-ev-deal-shazeer-character-ai-ceo-2024"],"attributes":{"role":"co-founder and CEO"}},{"id":"lg-cn-zhang-xiangyu-stepfun-chief-2026","statement":"36Kr, republishing a QbitAI report, identified Zhang Xiangyu on 27 January 2026 as StepFun’s chief scientist.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The record rests on reporting, not a company announcement; it does not date the start of the role. English rendering of the Chinese passage is a machine translation; no human translation review.","evidenceIds":["lg-ev-cn-zhang-xiangyu-stepfun-chief-2026"],"attributes":{"role":"首席科学家 (chief scientist)"}},{"id":"cmp-ms-fgf-2025-d6-competitor","statement":"Microsoft FGF · Feb 2025 states a provision on the comparison question “Provisions conditioned on other developers”, 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. A marginal-uplift consideration in the risk determination, measured against available tools and open-weights models; it does not state that measures change depending on other developers’ actions. 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-d6-competitor-1"],"attributes":{"provision":{"kind":"exception","key":"comparison-d6-competitor","facet":{"id":"D6.competitor","status":"partly"}}}},{"id":"ap-charles-i-jones-aghion-jones-jones-2017-ai-and-economic-growth","statement":"“Artificial Intelligence and Economic Growth” (2017-10) credits Charles I. Jones as one of its 3 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-charles-i-jones-aghion-jones-jones-2017-ai-and-economic-growth"]},{"id":"ap-eliezer-yudkowsky-sl4-2005-07-13-yudkowsky-0507-11477","statement":"The SL4 message “Re: Fighting UFAI” of 2005-07-13, archive message 11477, 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-sl4-2005-07-13-yudkowsky-0507-11477"]},{"id":"ap-eliezer-yudkowsky-lw-2025-re-recent-anthropic-safety-research","statement":"“Re: recent Anthropic safety research” (2025-08-06) credits Eliezer Yudkowsky 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-eliezer-yudkowsky-lw-2025-re-recent-anthropic-safety-research"]},{"id":"cmp-rsp-v3-4-d5-cadence","statement":"RSP · v3.4 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. Its “Off-cycle updates” analyse an individual model when a publicly deployed model is significantly more capable than all models whose relevant risks have been publicly analysed, or within 30 days for qualifying internal models. Each Risk Report covers the position as of a coverage date no more than 30 days before publication. 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-v3-4-d5-cadence-1"],"attributes":{"provision":{"kind":"evaluation_timing","key":"comparison-d5-cadence","facet":{"id":"D5.cadence","status":"partly"}}}},{"id":"c2-baidu-ernie-3-titan-report","statement":"Authors at Baidu and Peng Cheng Laboratory posted the ERNIE 3.0 Titan paper to arXiv on 23 December 2021 (version 1); Baidu's 20-F filed in March 2024 dates the release of ERNIE 3.0 Titan to 2022.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Most authors are marked with Baidu Inc. and four with Peng Cheng Laboratory; the paper’s credit to the laboratory is recorded separately and implies nothing about direction or control. The paper’s statements about model scale and performance are not recorded.","evidenceIds":["c2-ev-baidu-ernie-3-titan-report","c2-ev-baidu-ernie-3-titan-20f"]},{"id":"c2-moonshot-k15-api-access","statement":"The Kimi-k1.5 repository README, as of 22 January 2025, offered testing of Kimi k1.5 through the Kimi OpenPlatform API on application, and published no model weights.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The README describes API test access via an application form; the absence of weight files is read from the README and repository at this commit only, not from a statement that weights would never be released.","evidenceIds":["c2-ev-moonshot-k15-api-access"]},{"id":"c2-moonshot-k2-licence","statement":"Moonshot AI’s Kimi-K2 repository states that both the code repository and the model weights are released under a licence headed “Modified MIT License”; the README and the LICENSE file are present at the repository’s initial commit of 11 July 2025.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The licence is described as the text describes itself; no judgement is made whether it meets any definition of open source. It covers the repository named; other Kimi models carry their own licence files. Recorded until 2026-10-29 as a publication record of the publisher; since then, as what the document states (schema 2.20).","evidenceIds":["c2-ev-moonshot-k2-licence","c2-ev-moonshot-k2-licence-copyright","c2-ev-moonshot-k2-licence-2"],"attributes":{"provision":{"kind":"statement","key":"kimi-k2-licence-terms"}}},{"id":"c2-moonshot-k3-licence-maas","statement":"The Kimi K3 License states that if the licensee or any of its affiliates operates a “Model as a Service” business, as the licence defines it, and the aggregate revenue of the licensee and its affiliates exceeds 20 million US dollars (or the equivalent) over any consecutive 12 months, the licensee must enter into a separate agreement with Moonshot AI before using the software or its derivative works for any commercial purpose.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"What the licence text states. Section 2 defines Model as a Service as giving a third party access to model inference or fine-tuning with meaningful control over inputs, parameters or training data, and excludes end-user products whose model capabilities are embedded in specific features or harnesses, and the mere relaying of requests to models hosted by others. Section 4 exempts internal use, and use through Moonshot AI’s official products or certified inference partners, from Sections 2 and 3. The condition is not one on every commercial user. Whether it applies to any named business, and enforcement, are not recorded. Recorded until 2026-10-29 as a publication record of the publisher; since then, as what the document states (schema 2.20).","evidenceIds":["c2-ev-moonshot-k3-licence-maas","c2-ev-moonshot-k3-licence-maas-trigger"],"attributes":{"provision":{"kind":"statement","key":"kimi-k3-licence-maas"}}},{"id":"fl-cn-filings-2025-count","statement":"A CAC announcement reposted on gov.cn on 10 January 2025 states that, as of 31 December 2024, 302 generative AI services had completed filing with the CAC, 238 of them added in 2024, and that 105 applications or functions calling filed models had registered with local cyberspace offices in 2024.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Counts as announced by the CAC; not checked against the attached lists. No named model is recorded.","evidenceIds":["evidence-fl-cn-filings-2025-count-1","evidence-fl-cn-filings-2025-count-2"]},{"id":"ab-ev-llama3-risk-assessment","statement":"The Llama 3 paper states that, after validating its uplift results with CBRNE subject matter experts, it assesses that there is a low risk that release of the Llama 3 models will increase ecosystem risk related to biological or chemical weapon attacks.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"An assessment by the developers of the models released, based on their own study; it is recorded as their statement, not adopted. It concerns the models released in 2024 and is not evidence about later models or about weights released afterwards.","evidenceIds":["evidence-ab-ev-llama3-risk-assessment-1"],"attributes":{"evaluatedSystem":"The Llama 3 models as released in 2024","evaluation":{"what":"Whether releasing the models would increase ecosystem risk related to biological or chemical weapon attacks","how":"The developers' own reading of their uplift study, validated with subject matter experts","result":"An assessment of low risk that release will increase ecosystem risk","caveats":"An assessment by the releasing party, resting on the one study reported in the paper."}}},{"id":"ab-ev-claude-sonnet45-asl3-determination","statement":"The Claude Sonnet 4.5 system card states that the model demonstrated enhanced performance on several biological risk evaluations while remaining below ASL-4 rule-out thresholds, and that Anthropic has not determined whether it passed the ASL-3 capabilities threshold but cannot clearly rule out ASL-3 risks, so deploys it with ASL-3 measures as a precautionary, provisional action.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A developer's determination under its own policy, for the model and the automated evaluations the card describes. It is not an independent finding, and the comparison is to the developer's own earlier models. The card's biological risk results sections hold technical content; only the determination-level statements are cited.","evidenceIds":["evidence-ab-ev-claude-sonnet45-asl3-determination-1","evidence-ab-ev-claude-sonnet45-asl3-determination-2","evidence-ab-ev-claude-sonnet45-asl3-determination-3"],"attributes":{"evaluatedSystem":"Claude Sonnet 4.5, compared in the card with Claude Sonnet 4, Claude Opus 4 and Claude Opus 4.1","evaluation":{"what":"Whether the model reaches the ASL-3 or ASL-4 capability thresholds, CBRN among the domains, with biological risks the primary focus","how":"Comprehensive automated evaluations across the ASL-3 and ASL-4 thresholds, with comparative assessment against the developer's earlier models","result":"In the card's own words at the cited locator, the model remained well below ASL-4 thresholds across all domains of concern, and demonstrated enhanced performance on several biological risk evaluations while remaining below ASL-4 rule-out thresholds; deployment under ASL-3 protections","caveats":"The card states that it has not determined whether the model definitively passed the ASL-3 capabilities threshold, only that ASL-3 risks cannot clearly be ruled out, and that the ASL-3 measures are precautionary and provisional."}}}],"sources":[{"id":"source-learned-optimization","title":"Risks from Learned Optimization in Advanced Machine Learning Systems","authors":["Evan Hubinger","Chris van Merwijk","Vladimir Mikulik","Joar Skalse","Scott Garrabrant"],"publisher":"arXiv","publicationDate":"2019-06-05","version":"arXiv:1906.01820v3; claims scoped to this version, not first use in 2019","versionDate":"2021-12-01","canonicalUrl":"https://arxiv.org/abs/1906.01820","evidenceUrl":"https://arxiv.org/html/1906.01820v3","language":"en","accessedAt":"2026-09-17","reuse":{"attribution":"arXiv, Risks from Learned Optimization in Advanced Machine Learning Systems (2019-06-05).","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and minimal attributed excerpts only. Recheck terms before wider redistribution."}},{"id":"source-rsp-archive","title":"Responsible Scaling Policy — official version archive","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Mutable archive inspected 2026-09-18; version 3.4 effective July 8, 2026. Page update date is not the policy effective date.","canonicalUrl":"https://www.anthropic.com/responsible-scaling-policy","evidenceUrl":"https://www.anthropic.com/responsible-scaling-policy","language":"en","archivedUrl":"https://web.archive.org/web/20260918004547/https://www.anthropic.com/responsible-scaling-policy","archivedAt":"2026-09-18","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — official version archive.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-rsp-3-1","title":"Responsible Scaling Policy — v3.1","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Responsible Scaling Policy v3.1; effective date 2026-04-02.","versionDate":"2026-04-02","canonicalUrl":"https://www-cdn.anthropic.com/files/4zrzovbb/website/bf04581e4f329735fd90634f6a1962c13c0bd351.pdf","evidenceUrl":"https://www-cdn.anthropic.com/files/4zrzovbb/website/bf04581e4f329735fd90634f6a1962c13c0bd351.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260407193305/https://www-cdn.anthropic.com/files/4zrzovbb/website/bf04581e4f329735fd90634f6a1962c13c0bd351.pdf","archivedAt":"2026-04-07","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v3.1.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-rsp-3-4","title":"Responsible Scaling Policy — v3.4","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Responsible Scaling Policy v3.4; effective date 2026-07-08.","versionDate":"2026-07-08","canonicalUrl":"https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf","evidenceUrl":"https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260710234036/https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf","archivedAt":"2026-07-10","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v3.4.","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-leadership-2025","title":"Safe Superintelligence Inc. — Updates","authors":["Ilya Sutskever","Safe Superintelligence Inc."],"publisher":"Safe Superintelligence Inc.","publicationDate":"2025-07-03","version":"Exact July 3, 2025 signed message retained on a mutable multi-date updates page. Later updates are outside these claims.","canonicalUrl":"https://ssi.inc/updates","evidenceUrl":"https://ssi.inc/updates","language":"en","archivedUrl":"https://web.archive.org/web/20260903031538/https://ssi.inc/updates","archivedAt":"2026-09-03","accessedAt":"2026-09-19","reuse":{"attribution":"Ilya Sutskever; Safe Superintelligence Inc., Safe Superintelligence Inc. — Updates (2025-07-03)","licence":"No open redistribution licence asserted","note":"Bibliographic metadata and brief attributed paraphrases only; no full source or personal contact details redistributed."}},{"id":"source-atlas-openai-lp","title":"OpenAI LP","authors":["Greg Brockman","Ilya Sutskever","OpenAI"],"publisher":"OpenAI","publicationDate":"2019-03-11","version":"Dated publisher page inspected 2026-09-28 and again 2026-10-04 at its current address; the quotes used from 2026-10-04 also match the Internet Archive capture of 2019-03-11 (16:15 UTC) at the original address, which uses straight quotation marks and an additional standfirst.","canonicalUrl":"https://openai.com/index/openai-lp/","evidenceUrl":"https://openai.com/index/openai-lp/","language":"en","archivedUrl":"https://web.archive.org/web/20190311161512/https://openai.com/blog/openai-lp/","archivedAt":"2019-03-11","accessedAt":"2026-10-04","reuse":{"attribution":"Greg Brockman; Ilya Sutskever; OpenAI. OpenAI LP.","licence":"No redistribution licence asserted.","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-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-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-michie-1986-on-machine-intelligence","title":"On Machine Intelligence, second edition (chapter 20, \"Computer—servant or master\")","authors":["Donald Michie"],"publisher":"Ellis Horwood","publicationDate":"1986","version":"Scan of the 1986 second edition hosted by gwern.net; imprint page and chapter 20 read by the pre-1996 candidate search.","canonicalUrl":"https://gwern.net/doc/ai/1986-michie-onmachineintelligence.pdf","evidenceUrl":"https://gwern.net/doc/ai/1986-michie-onmachineintelligence.pdf","language":"en","accessedAt":"2026-10-06","reuse":{"attribution":"Donald Michie, On Machine Intelligence, 2nd edn (Chichester: Ellis Horwood, 1986), chapter 20.","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-logi-miri-reprint","title":"Levels of Organization in General Intelligence (MIRI reprint)","authors":["Eliezer Yudkowsky"],"publisher":"Machine Intelligence Research Institute","publicationDate":"2007","version":"MIRI reprint, \"This version contains minor changes\", retrieved 2026-10-06 with the hash recorded by the 1996–2022 candidate search.","canonicalUrl":"https://intelligence.org/files/LOGI.pdf","evidenceUrl":"https://intelligence.org/files/LOGI.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20130816132711/http://intelligence.org/files/LOGI.pdf","archivedAt":"2013-08-16","accessedAt":"2026-10-06","reuse":{"attribution":"Eliezer Yudkowsky, \"Levels of Organization in General Intelligence\", in Artificial General Intelligence (2007); MIRI reprint.","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-2025-re-recent-anthropic-safety-research","title":"Re: recent Anthropic safety research","authors":["Eliezer Yudkowsky"],"publisher":"LessWrong","publicationDate":"2025-08-06","version":"Post body retrieved through the LessWrong GraphQL API on 2026-10-06 (SHA-256 6d9a16d1be287ce9ff3041b28bef6aa2807eefab2d09c56e3fb1cef49a9f7710), content version 1.7.0; quotations also compared with the posting-day version unavailable from the API.","canonicalUrl":"https://www.lesswrong.com/posts/oDX5vcDTEei8WuoBx/re-recent-anthropic-safety-research","evidenceUrl":"https://www.lesswrong.com/posts/oDX5vcDTEei8WuoBx/re-recent-anthropic-safety-research","language":"en","archivedUrl":"https://web.archive.org/web/20260921070433/https://www.lesswrong.com/posts/oDX5vcDTEei8WuoBx/re-recent-anthropic-safety-research","archivedAt":"2026-09-21","accessedAt":"2026-10-06","reuse":{"attribution":"Eliezer Yudkowsky, “Re: recent Anthropic safety research”, LessWrong, 2025-08-06.","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-sl4-2005-07-13-yudkowsky-0507-11477","title":"Re: Fighting UFAI (SL4 mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"SL4 mailing list","publicationDate":"2005-07-13","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 1ba0e5933533210210b176aca4c4ced8e0ad4f996ad71908d77ac7e6c369aaf3); 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/0507/11477.html","evidenceUrl":"http://sl4.org/archive/0507/11477.html","language":"en","archivedUrl":"https://web.archive.org/web/20251115083834/http://sl4.org/archive/0507/11477.html","archivedAt":"2025-11-15","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Re: Fighting UFAI”, SL4 mailing list, 2005-07-13.","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-eu-gpai-guidelines-2025","title":"Communication from the Commission: Commission Guidelines on the scope of the obligations for providers of general-purpose AI models established by Regulation (EU) 2024/1689 (AI Act), C(2025) 7719 final","authors":["European Commission"],"publisher":"European Commission","publicationDate":"2025-11-19","version":"PDF dated Brussels 19.11.2025, retrieved 2026-10-11 from the Commission download address (SHA-1 1ecfd7b62bf0fbb659ab390bb15a2ad2ad71af13). The library entry for the guidelines gives Publication 18 July 2025; the text of the July version was not retrieved or compared. No archive capture is cited: the Internet Archive refused lookup requests (HTTP 429) on 2026-10-11, so none was checked.","canonicalUrl":"https://ec.europa.eu/newsroom/dae/redirection/document/118340","evidenceUrl":"https://ec.europa.eu/newsroom/dae/redirection/document/118340","language":"en","archivedUrl":"https://web.archive.org/web/20261009015109/https://ec.europa.eu/newsroom/dae/redirection/document/118340","archivedAt":"2026-10-09","accessedAt":"2026-10-11","reuse":{"attribution":"European Commission, Commission Guidelines on the scope of the obligations for providers of general-purpose AI models, C(2025) 7719 final (19 November 2025).","licence":"Official 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-aghion-jones-jones-2017-nber-capture","title":"Artificial Intelligence and Economic Growth","authors":["Philippe Aghion","Benjamin F. Jones","Charles I. Jones"],"publisher":"NBER Working Paper 23928","publicationDate":"2017-10","version":"Wayback capture of the NBER PDF (w23928); compared with the author-hosted Version 1.0 (web.stanford.edu/~chadj/AI.pdf, 2018 capture). Capture retrieved 2026-10-08 (SHA-256 a8e68273c738c80fe565afa60bc27ec5e8aaf3135011a0f760233896f1f2100c). Versions: NBER Working Paper 23928 (2017-10): checked; author copy (Stanford), version 1.0 (2017-10-10): compared; book chapter, University of Chicago Press (2019): not retrieved. Published version: Aghion, P., Jones, B. F., & Jones, C. I. (2019). Artificial intelligence and economic growth. In A. Agrawal, J. Gans, & A. Goldfarb (Eds.), The economics of artificial intelligence: An agenda (pp. 237–290). University of Chicago Press. https://doi.org/10.7208/chicago/9780226613475.003.0009 Working paper: NBER Working Paper No. 23928 (2017). https://doi.org/10.3386/w23928. Re-inspected: Full text. Current copy retrieved directly on 2026-10-09; no archive capture (Wayback unreachable), so earlier bytes are not authenticated. SHA-256 a8e68273c738c80fe565afa60bc27ec5e8aaf3135011a0f760233896f1f2100c. (Merged on 2026-10-29 with the record of the re-inspection, source-aghion-jones-jones-2017-nber, by consistency audit fix B4.)","canonicalUrl":"https://www.nber.org/system/files/working_papers/w23928/w23928.pdf","evidenceUrl":"https://web.archive.org/web/20210303021500/https://www.nber.org/system/files/working_papers/w23928/w23928.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20210303021500/https://www.nber.org/system/files/working_papers/w23928/w23928.pdf","archivedAt":"2021-03-03","accessedAt":"2026-10-09","reuse":{"attribution":"Philippe Aghion, Benjamin F. Jones, Charles I. Jones, “Artificial Intelligence and Economic Growth”, NBER Working Paper 23928, 2017-10.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-leontief-duchin-1984-impacts-of-automation","title":"The Impacts of Automation on Employment, 1963–2000. Final Report","authors":["Wassily Leontief","Faye Duchin"],"publisher":"New York: Institute for Economic Analysis, New York University; National Science Foundation contract PRA-8012844; April 1984, 376 pp.; ERIC ED 241 743","publicationDate":"1984-04","version":"Internet Archive OCR text (djvu.txt) of a scanned copy, retrieved 2026-10-09 from https://dn711508.ca.archive.org/0/items/microIA411534480031/microIA411534480031djvu.txt; OCR errors possible; quotes checked only against clean stretches. Current copy; no earlier capture compared. SHA-256 cd03df677ca1a723f843eccec0fa1415b3227d604bc83c6783a40321b9a785e9. Book version not retrieved; the 1984 report was. Leontief, W., & Duchin, F. (1986). The future impact of automation on workers. Oxford University Press.","canonicalUrl":"https://archive.org/details/micro_IA41153448_0031","evidenceUrl":"https://archive.org/details/micro_IA41153448_0031","language":"en","accessedAt":"2026-10-09","reuse":{"attribution":"Wassily Leontief, Faye Duchin, “The Impacts of Automation on Employment, 1963–2000. Final Report”, New York: Institute for Economic Analysis, New York University; National Science Foundation contract PRA-8012844; April 1984, 376 pp.; ERIC ED 241 743, 1984-04.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"lg-src-cn-qbitai-stepfun-2026","title":"阶跃星辰不再低调：巨额融资，印奇加入，“1+3”核心决策层浮出水面","authors":["量子位 (QbitAI)"],"publisher":"36氪 (36Kr), republishing 量子位","publicationDate":"2026-01-27","version":"Dated page, retrieved on 2026-10-15 (SHA-256 8081f6283cd0326b9aeb5becc09aa65187d5df211d0d94ab82444c68c7631473) directly from the publisher.","canonicalUrl":"https://www.36kr.com/p/3657567182938757","evidenceUrl":"https://www.36kr.com/p/3657567182938757","language":"zh","archivedUrl":"https://web.archive.org/web/20261029003800/https://www.36kr.com/p/3657567182938757","archivedAt":"2026-10-29","accessedAt":"2026-10-15","reuse":{"attribution":"量子位 (QbitAI), “阶跃星辰不再低调：巨额融资，印奇加入，“1+3”核心决策层浮出水面”, 2026-01-27.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"lg-src-deal-tc-cai-google-2024","title":"Character.AI CEO Noam Shazeer returns to Google","authors":["Ivan Mehta"],"publisher":"TechCrunch","publicationDate":"2024-08-02","version":"Dated page, retrieved on 2026-10-15 (SHA-256 b9dd741f85ef8319d9d767ff1d6e5805da3702f775d2c3cbade85fb292412b22) directly from the publisher.","canonicalUrl":"https://techcrunch.com/2024/08/02/character-ai-ceo-noam-shazeer-returns-to-google/","evidenceUrl":"https://techcrunch.com/2024/08/02/character-ai-ceo-noam-shazeer-returns-to-google/","language":"en","archivedUrl":"https://web.archive.org/web/20261015012331/https://techcrunch.com/2024/08/02/character-ai-ceo-noam-shazeer-returns-to-google/","archivedAt":"2026-10-15","accessedAt":"2026-10-15","reuse":{"attribution":"Ivan Mehta, “Character.AI CEO Noam Shazeer returns to Google”, 2024-08-02.","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":"c2-src-baidu-20f-fy2023","title":"Baidu, Inc. annual report on Form 20-F for the fiscal year ended December 31, 2023","authors":["Baidu, Inc."],"publisher":"U.S. Securities and Exchange Commission (EDGAR)","publicationDate":"2024-03-15","version":"Dated document, retrieved on 2026-10-27 (SHA-256 745a5397f9149a6d957a64226a03cc382735a41e5b81772786daa538ddd882a3) through the third-party reader relay r.jina.ai, which renders the publisher’s current page (direct retrieval failed); not an archive copy. Version: 0001193125-24-068527. Filed document d584913d20f.htm, filing date 15 March 2024; retrieved through the r.jina.ai relay after direct retrieval failed (HTTP/2 stream errors).","canonicalUrl":"https://www.sec.gov/Archives/edgar/data/1329099/000119312524068527/d584913d20f.htm","evidenceUrl":"https://www.sec.gov/Archives/edgar/data/1329099/000119312524068527/d584913d20f.htm","language":"en","accessedAt":"2026-10-27","reuse":{"attribution":"Baidu, Inc., “Baidu, Inc. annual report on Form 20-F for the fiscal year ended December 31, 2023”, 2024-03-15.","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-ernie3-titan-arxiv-v1","title":"ERNIE 3.0 Titan: Exploring Larger-scale Knowledge Enhanced Pre-training for Language Understanding and Generation","authors":["Shuohuan Wang","Yu Sun","Yang Xiang"],"publisher":"arXiv","publicationDate":"2021-12-23","version":"Dated document, retrieved on 2026-10-27 (SHA-256 cb22312dce198f418286264665a60f22d5f09dc9335080d246b0605ac4d8d0a4) directly from the publisher. Version: 2112.12731v1. Version 1 as submitted on 23 December 2021. The author list gives the first credited authors only; see the document for the full list.","canonicalUrl":"https://arxiv.org/pdf/2112.12731v1","evidenceUrl":"https://arxiv.org/pdf/2112.12731v1","language":"en","accessedAt":"2026-10-27","reuse":{"attribution":"Shuohuan Wang, Yu Sun, Yang Xiang, “ERNIE 3.0 Titan: Exploring Larger-scale Knowledge Enhanced Pre-training for Language Understanding and Generation”, 2021-12-23.","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":"c2-src-kimi-k15-readme","title":"Kimi-k1.5 repository README","authors":["Kimi Team"],"publisher":"MoonshotAI (GitHub)","publicationDate":"2025-01-22","version":"Dated page, retrieved on 2026-10-27 (SHA-256 95f1751aebf2ed76bade008bfe016afaf374f34d6a2e930412c3dbf58eb767f5) directly from the publisher. Version: 617e7c2d80970eb7c52c6cd694b353e6a98884dd. README at the commit of 22 January 2025 (\"Update README.md and Report\"); the repository was initialised on 20 January 2025; later edits not inspected.","canonicalUrl":"https://raw.githubusercontent.com/MoonshotAI/Kimi-k1.5/617e7c2d80970eb7c52c6cd694b353e6a98884dd/README.md","evidenceUrl":"https://raw.githubusercontent.com/MoonshotAI/Kimi-k1.5/617e7c2d80970eb7c52c6cd694b353e6a98884dd/README.md","language":"en","archivedUrl":"https://web.archive.org/web/20261028220806/https://raw.githubusercontent.com/MoonshotAI/Kimi-k1.5/617e7c2d80970eb7c52c6cd694b353e6a98884dd/README.md","archivedAt":"2026-10-28","accessedAt":"2026-10-27","reuse":{"attribution":"Kimi Team, “Kimi-k1.5 repository README”, 2025-01-22.","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":"c2-src-kimi-k2-license","title":"Kimi-K2 LICENSE (Modified MIT License)","authors":["Moonshot AI"],"publisher":"MoonshotAI (GitHub)","publicationDate":"2025-07-11","version":"Dated document, retrieved on 2026-10-27 (SHA-256 b280da6f496d75e0c5beb79b03520d190fce67dfe12f75cf3aadad8670839d93) directly from the publisher. Version: d778e270ff3551ef3e42a085dd474760501a95b7. LICENSE file at the initial commit of 11 July 2025; the file history shows no later change in the inspected clone.","canonicalUrl":"https://raw.githubusercontent.com/MoonshotAI/Kimi-K2/d778e270ff3551ef3e42a085dd474760501a95b7/LICENSE","evidenceUrl":"https://raw.githubusercontent.com/MoonshotAI/Kimi-K2/d778e270ff3551ef3e42a085dd474760501a95b7/LICENSE","language":"en","archivedUrl":"https://web.archive.org/web/20261028220827/https://raw.githubusercontent.com/MoonshotAI/Kimi-K2/d778e270ff3551ef3e42a085dd474760501a95b7/LICENSE","archivedAt":"2026-10-28","accessedAt":"2026-10-27","reuse":{"attribution":"Moonshot AI, “Kimi-K2 LICENSE (Modified MIT License)”, 2025-07-11.","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-kimi-k2-readme-0711","title":"Kimi-K2 repository README","authors":["Moonshot AI"],"publisher":"MoonshotAI (GitHub)","publicationDate":"2025-07-11","version":"Dated page, retrieved on 2026-10-27 (SHA-256 24569c90a1986572fdd626d6527149bfe86561f6fa2512412c33401f772ab676) directly from the publisher. Version: d778e270ff3551ef3e42a085dd474760501a95b7. README at the initial commit of 11 July 2025; later edits not inspected.","canonicalUrl":"https://raw.githubusercontent.com/MoonshotAI/Kimi-K2/d778e270ff3551ef3e42a085dd474760501a95b7/README.md","evidenceUrl":"https://raw.githubusercontent.com/MoonshotAI/Kimi-K2/d778e270ff3551ef3e42a085dd474760501a95b7/README.md","language":"en","archivedUrl":"https://web.archive.org/web/20261028223606/https://raw.githubusercontent.com/MoonshotAI/Kimi-K2/d778e270ff3551ef3e42a085dd474760501a95b7/README.md","archivedAt":"2026-10-28","accessedAt":"2026-10-27","reuse":{"attribution":"Moonshot AI, “Kimi-K2 repository README”, 2025-07-11.","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-kimi-k3-license","title":"Kimi-K3 LICENSE (Kimi K3 License)","authors":["Moonshot AI"],"publisher":"moonshotai (Hugging Face)","publicationDate":"2026-07-27","version":"Dated document, retrieved on 2026-10-27 (SHA-256 20c797ce19af0c17de52c6afb144644768a591c521655f5ebf5712c9850f2887) directly from the publisher. Version: 2496450e92e425c886db095102a52a6682ca3970. LICENSE at the Hugging Face commit of 27 July 2026 15:14 UTC; later edits not inspected.","canonicalUrl":"https://huggingface.co/moonshotai/Kimi-K3/raw/2496450e92e425c886db095102a52a6682ca3970/LICENSE","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K3/raw/2496450e92e425c886db095102a52a6682ca3970/LICENSE","language":"en","archivedUrl":"https://web.archive.org/web/20261028223741/https://huggingface.co/moonshotai/Kimi-K3/raw/2496450e92e425c886db095102a52a6682ca3970/LICENSE","archivedAt":"2026-10-28","accessedAt":"2026-10-27","reuse":{"attribution":"Moonshot AI, “Kimi-K3 LICENSE (Kimi K3 License)”, 2026-07-27.","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-cn-cac-genai-filings-2025","title":"国家互联网信息办公室关于发布2024年生成式人工智能服务已备案信息的公告 (CAC announcement publishing 2024 information on filed generative AI services)","authors":["国家互联网信息办公室 (Cyberspace Administration of China)"],"publisher":"中国政府网 (www.gov.cn), 部门动态 (source: 国家网信办网站)","publicationDate":"2025-01-10","version":"Retrieved on 2026-10-27 (SHA-256 61a16fdea3255d5c99a5582ecc77277a4133de2814239b53e78ea6427c795ebc). gov.cn department-news page dated 2025-01-10 10:16, retrieved 2026-10-27. The attachments were not retrieved. Its wording can change; earlier wording is not authenticated. The English in parentheses in the title is a machine translation.","canonicalUrl":"https://www.gov.cn/lianbo/bumen/202501/content_7002186.htm","evidenceUrl":"https://www.gov.cn/lianbo/bumen/202501/content_7002186.htm","language":"zh-Hans","accessedAt":"2026-10-27","reuse":{"attribution":"国家互联网信息办公室 (Cyberspace Administration of China), “国家互联网信息办公室关于发布2024年生成式人工智能服务已备案信息的公告 (CAC announcement publishing 2024 information on filed generative AI services)”, 2025-01-10.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-ab-claude-sonnet45-card","title":"System Card: Claude Sonnet 4.5","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 d5d1ce28fea101ae3b201a5d4da100090a5d6b73246f7da820cafd7dbb11100e). PDF whose cover reads 'System Card: Claude Sonnet 4.5, September 2025', as served on 2026-10-28; the request redirected to https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf. No day of the month is stated. Retrieved again at admission on 2026-10-30 (SHA-256 d5d1ce28fea101ae3b201a5d4da100090a5d6b73246f7da820cafd7dbb11100e, the same bytes).","canonicalUrl":"https://assets.anthropic.com/m/12f214efcc2f457a/original/Claude-Sonnet-4-5-System-Card.pdf","evidenceUrl":"https://assets.anthropic.com/m/12f214efcc2f457a/original/Claude-Sonnet-4-5-System-Card.pdf","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"Anthropic, “System Card: Claude Sonnet 4.5”.","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-llama3-paper","title":"The Llama 3 Herd of Models (arXiv:2407.21783v3)","authors":["arXiv (preprint posted by the Llama 3 authors, Meta)"],"publisher":"arXiv (preprint posted by the Llama 3 authors, Meta)","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 481f1599468f95a07d05e97fafe55bbe786dc1624c6f881bcb4c7d14c933d083). Version 3 of the preprint (the version served at /pdf/2407.21783v3 on 2026-10-28). Retrieved again at admission on 2026-10-30 (SHA-256 481f1599468f95a07d05e97fafe55bbe786dc1624c6f881bcb4c7d14c933d083, the same bytes).","canonicalUrl":"https://arxiv.org/pdf/2407.21783v3","evidenceUrl":"https://arxiv.org/pdf/2407.21783v3","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"arXiv (preprint posted by the Llama 3 authors, Meta), “The Llama 3 Herd of Models (arXiv:2407.21783v3)”.","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-risks-discusses-deception-2021","sourceRole":"primary","support":"supports","locator":"Section 4, Deceptive alignment, opening paragraphs","passage":"Section 4 describes a mesa-optimizer pursuing the base objective instrumentally in order to avoid modification.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-17","note":"Source passage inspected by an automated research assistant. Human editorial review is pending."},{"id":"evidence-supersedes-rsp-3-1","sourceRole":"primary","support":"supports","locator":"PDF p. 20, Changelog, entry for 2026-04-02; official archive orders current and previous policy versions","passage":"The dated changelog places the April 2, 2026 (RSP v3.1) entry directly after the February 24, 2026 (RSP v3.0) entry; the v3.1 entry does not itself name the version it updates.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-03","note":"Paraphrase re-checked by an automated research assistant on 2026-10-03 against the inspected PDF, byte-identical to the Internet Archive capture of 2026-04-07 and reworded so that it does not imply the entry names its predecessor. Human review pending."},{"id":"evidence-supersedes-rsp-3-1-2","sourceRole":"primary","support":"supports","locator":"Policy versions, adjacent v3.1 and v3.0 entries","passage":"The publisher lists these successive editions in its policy archive.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-18","note":"Official source inspected by machine; passage is an editorial paraphrase. Human review pending."},{"id":"evidence-gross-ssi-departure-announced-2025","sourceRole":"primary","support":"supports","locator":"July 3rd, 2025 update, first paragraph","passage":"SSI reports Gross’s separation effective June 29 in its July 3 announcement.","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-cm-seoul-g42-1","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 “G42”.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-04","note":"Editorial paraphrase, re-checked on 2026-10-04 by the integrating session against its own direct fetch of the cited expression. The list is identical in the Internet Archive capture of 2024-05-21. The integrating session wrote or selected this passage from its own reading of the source; it does not repeat a passage from the week 4b research pass. Human editorial review is pending."},{"id":"evidence-ts-b6-michie-1968-contests","sourceRole":"primary","support":"supports","locator":"Chapter 20, end of printed p. 244 to \"Helpers and Hobbies\" on p. 246 (PDF pages 249–251)","passage":"Michie expects computers to function as helpers in a division of labour with people, not as masters or servants.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-06","note":"Paraphrase from the pre-1996 candidate search."},{"id":"evidence-ts-a2-logi","sourceRole":"primary","support":"supports","locator":"Section 3 \"Seed AI\", introduction, printed p. 96; section 3.2 \"Recursive Self-enhancement\", printed p. 102 (PDF pp. 97, 103)","passage":"A seed AI is an AI designed for self-understanding, self-modification, and recursive self-improvement. [...] Recursive self-improvement is a consequence of seed AI, not a cheap way to achieve AI.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-06","note":"Checked against the reprint’s text layer on 2026-10-06."},{"id":"evidence-eugl-non-binding-1","sourceRole":"primary","support":"supports","locator":"Paragraph (9), first sentence","passage":"These guidelines are not binding for providers of general-purpose AI models; an authoritative interpretation of the AI Act may only be given by the Court of Justice of the European Union","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-11","note":"Verbatim quote, matched mechanically (after whitespace normalisation) by the proposing machine agent against the text extracted from the file retrieved on 2026-10-11. Checked by fresh-context machine audit on 2026-10-11 (review/frontier-governance-2-audit-2026-10-11.md). Human editorial review is pending."},{"id":"evidence-eugl-non-binding-2","sourceRole":"primary","support":"supports","locator":"Paragraph (9), second sentence","passage":"these guidelines set out the Commission’s interpretation and application of the AI Act, on which it will base its enforcement action","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-29","note":"Verbatim quote, matched mechanically (after whitespace normalisation) by the proposing machine agent against the text extracted from the file retrieved on 2026-10-11. Checked by fresh-context machine audit on 2026-10-11 (review/frontier-governance-2-audit-2026-10-11.md). Human editorial review is pending. On 2026-10-30, by the maintainer’s decision of that day, the quotation was corrected to the archive capture of 2026-10-09 (quotation marks and apostrophes as in the capture), which a cross-family archive check read on 2026-10-29; the proposition is unchanged."},{"id":"evidence-ec-m08-01","sourceRole":"primary","support":"supports","locator":"Abstract ('Findings'), unnumbered front matter","passage":"The intensive use of automation over the next twenty years will make it possible to conserve about 10% of the labor that would have been required to produce the same bills of goods in the absence of increased automation.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-14","note":"Candidate pack passage M08-01 (research/economics-mainstream). The quotation (39 words) matched the retrieved current copy (SHA-256 cd03df677ca1a723f843eccec0fa1415b3227d604bc83c6783a40321b9a785e9) in the pack's verification. Re-checked at admission on 2026-10-14 against a copy retrieved that day.","contextOnly":{"reason":"Fresh-context audit (A2) judged that the quotation misleads when shown alone.","requires":"A model-scenario comparison (dynamic input-output scenarios S1-S4), not an unconditional forecast. 'Same bills of goods' means a counterfactual at fixed final demand."}},{"id":"lg-ev-oai-shivon-zilis-nonprofit-board-2019","sourceRole":"primary","support":"supports","locator":"Section “Who’s involved”, board list (footnote A: board as of 11 March 2019)","passage":"The board list names Shivon Zilis among the non-employee members of OpenAI Nonprofit’s board.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The paraphrase’s verbatim anchor “Reid Hoffman, Shivon Zilis” 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. The paraphrase is Selekoa’s; only the anchor is the source’s wording. No human editorial review.","retrievalRoute":"reader_relay"},{"id":"lg-ev-oai-shivon-zilis-nonprofit-board-2019-inc","sourceRole":"primary","support":"context","locator":"Paragraph defining the general partner","passage":"OpenAI’s 2019 note gives the legal name of OpenAI Nonprofit as OpenAI Inc.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The paraphrase’s verbatim anchor “whose legal name is “OpenAI Inc”” 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. The paraphrase is Selekoa’s; only the anchor is the source’s wording. No human editorial review.","retrievalRoute":"reader_relay"},{"id":"lg-ev-deal-shazeer-character-ai-ceo-2024","sourceRole":"secondary","support":"supports","locator":"Paragraph 1","passage":"Character.AI co-founder and CEO Noam Shazeer is returning to Google","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (10 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read directly from the publisher. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-16/17) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"lg-ev-cn-zhang-xiangyu-stepfun-chief-2026","sourceRole":"secondary","support":"supports","locator":"Paragraph listing the “1+3” group","passage":"分别是阶跃星辰CEO姜大昕，首席科学家张祥雨，CTO朱亦博","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (1 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read directly from the publisher. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-16/17) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"ev-cmp-ms-fgf-2025-d6-competitor-1","sourceRole":"primary","support":"supports","locator":"PDF p. 7 (printed p. 6), §3, Deeper capability assessment, 'Holistic risk assessment' bullet, final sentence","passage":"considers the marginal capability uplift … including currently available open-weights models.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-27","note":"Verbatim excerpt 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. Part of the holistic risk assessment whose results 'then inform a determination' of whether a model has a tracked capability and at what level. Human editorial review is pending."},{"id":"evidence-ap-charles-i-jones-aghion-jones-jones-2017-ai-and-economic-growth","sourceRole":"primary","support":"supports","locator":"PDF cover, beneath the title; repeated on the abstract page","passage":"Charles I. Jones","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, who corrected the quotation to the printed order. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"evidence-ap-eliezer-yudkowsky-sl4-2005-07-13-yudkowsky-0507-11477","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-2025-re-recent-anthropic-safety-research","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 and checked by a reviewer from a different model family on 2026-10-27. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"ev-cmp-rsp-v3-4-d5-cadence-1","sourceRole":"primary","support":"supports","locator":"PDF p. 11, §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":"c2-ev-baidu-ernie-3-titan-report","sourceRole":"primary","support":"supports","locator":"Title page and arXiv stamp","passage":"arXiv:2112.12731v1 [cs.CL] 23 Dec 2021","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 (5 words) 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. No human editorial review.","retrievalRoute":"direct"},{"id":"c2-ev-baidu-ernie-3-titan-20f","sourceRole":"primary","support":"context","locator":"Item 4.B, Business overview, 'AI Capabilities'","passage":"In 2022, Baidu released ERNIE 3.0 Titan, a pre-training language model with 260 billion parameters.","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 (15 words) matched the retrieved text on 2026-10-27 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 (direct retrieval failed); 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":"c2-ev-moonshot-k15-api-access","sourceRole":"primary","support":"supports","locator":"README, “Test Model by API”","passage":"You can test Kimi k1.5 through the Kimi OpenPlatform. Fill out the test application form in","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 (16 words) 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. No human editorial review.","retrievalRoute":"direct"},{"id":"c2-ev-moonshot-k2-licence","sourceRole":"primary","support":"supports","locator":"LICENSE, heading","passage":"Modified MIT License","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 (3 words), a part of a longer passage that the pack joined across separate fields of the page, 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. The pack had joined this field with others of the same page into one quotation; at admission each field became its own quotation (cross-family check, 2026-10-27). A cross-family machine check (2026-10-27, item K039) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"c2-ev-moonshot-k2-licence-copyright","sourceRole":"primary","support":"supports","locator":"LICENSE, copyright line","passage":"Copyright (c) 2025 Moonshot AI","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 (5 words), a part of a longer passage that the pack joined across separate fields of the page, 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. The pack had joined this field with others of the same page into one quotation; at admission each field became its own quotation (cross-family check, 2026-10-27). A cross-family machine check (2026-10-27, item K039) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"c2-ev-moonshot-k2-licence-2","sourceRole":"primary","support":"supports","locator":"README, “6. License”","passage":"Both the code repository and the model weights are released under the","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 (12 words) 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. A cross-family machine check (2026-10-27, item K039) re-read the record’s sources and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"c2-ev-moonshot-k3-licence-maas","sourceRole":"primary","support":"supports","locator":"LICENSE, Section 2, second paragraph, end","passage":"the Licensee must enter into a separate agreement with Moonshot AI","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 (11 words) 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. Shortened at admission to keep the quotations from this source within the per-document limit; the shortened wording is part of the checked passage. A cross-family machine check (2026-10-27, item K062) re-read the record’s sources and judged it partly supported; its corrections were applied at admission. No human editorial review.","retrievalRoute":"direct"},{"id":"c2-ev-moonshot-k3-licence-maas-trigger","sourceRole":"primary","support":"supports","locator":"LICENSE, Section 2, second paragraph, with the definition in its first paragraph and the exemptions of Section 4","passage":"Section 2 applies if the Licensee or any of its affiliates operates a Model as a Service business and the aggregate revenue of the Licensee and its affiliates exceeds 20 million US dollars, or the equivalent in other currencies, in total over any consecutive 12 months; Section 4 exempts internal use and use through Moonshot AI’s official products or certified inference partners.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-27","note":"Added at the admission of the Chinese frontier-AI institutions, second collection (research/china-labs-2). The cross-family check (K062) found that the pack’s record left out the affiliates and the parties whose revenue is aggregated. Section 2 and Section 4 were read at admission from a direct retrieval of the pinned file whose SHA-256 equals the pack’s retrieval. The anchor “aggregate revenue of the Licensee and its affiliates” was matched in that retrieval. The paraphrase is Selekoa’s; only the anchor is the source’s wording. A cross-family machine check (2026-10-27, item K062) re-read the record’s sources and judged it partly supported; its corrections were applied at admission. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-fl-cn-filings-2025-count-1","sourceRole":"primary","support":"supports","locator":"Announcement text, first paragraph","passage":"共302款生成式人工智能服务在国家网信办完成备案","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-28","note":"Candidate pack passage, frameworks to law (research/frameworks-to-law). The quotation (22 words, each CJK character counted as one) 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.gov.cn. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. English machine translation, not verified by a person: “a total of 302 generative AI services completed filing with the CAC”. A cross-family machine check (2026-10-28, item 71) answered the record’s question blind from fresh retrievals and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-fl-cn-filings-2025-count-2","sourceRole":"primary","support":"supports","locator":"Announcement text, first paragraph","passage":"The same sentence gives 238 filings added in 2024 and 105 applications or functions, calling filed models via API or otherwise, registered with local cyberspace offices in 2024; counts as of 31 December 2024.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-28","note":"Candidate pack passage, frameworks to law (research/frameworks-to-law). The paraphrase is Selekoa’s, written from the text retrieved on 2026-10-27 and read directly from www.gov.cn; it is not a quotation and was not machine-matched. A cross-family machine check (2026-10-28, item 71) answered the record’s question blind from fresh retrievals and agreed with the record. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-llama3-risk-assessment-1","sourceRole":"primary","support":"supports","locator":"PDF p. 48, §5.4.5, closing sentence","passage":"After validating these results with CBRNE SMEs, we assess that there is a low risk that release of Llama 3 models will increase ecosystem risk related to biological or chemical weapon attacks.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (32 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 22) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude-sonnet45-asl3-determination-1","sourceRole":"primary","support":"supports","locator":"PDF p. 11 (printed p. 10), §1.2.4 Conclusions, third bulleted conclusion","passage":"Demonstrated enhanced performance on several biological risk evaluations while remaining below ASL-4 rule-out thresholds","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. A cross-family machine check (2026-10-30, item 41) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude-sonnet45-asl3-determination-2","sourceRole":"primary","support":"supports","locator":"PDF p. 11 (printed p. 10), §1.2.4 Conclusions, first bulleted conclusion","passage":"Remained well below ASL-4 thresholds across all domains of concern","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (10 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. Added after the audit of 2026-10-28, which found the result attribute reading 'ASL-4 capabilities ruled out across all domains' where the cited bullets say only this. The stronger 'confidently rule out ASL-4 capabilities across all domains' appears at §9.1 (PDF p. 126), a locator this record does not cite, and is not recorded here. A cross-family machine check (2026-10-30, item 41) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude-sonnet45-asl3-determination-3","sourceRole":"primary","support":"context","locator":"PDF p. 11 (printed p. 10), §1.2.4 Conclusions, closing paragraph","passage":"we have not determined whether Claude Sonnet 4.5 has definitively passed the capabilities threshold that requires ASL-3 protections. Rather, we cannot clearly rule out ASL-3 risks for Claude Sonnet 4.5.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (30 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 41) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"}],"at":{"entities":[19,25,28,45,46,49,75,81,201,207,242,252,265,315,337,346,436,481,496,502,521,527,579,635,644,747,755,797,803,840,889,948,960,962,969,1014,1237,1242],"claims":[25,78,180,704,796,923,1218,1445,1603,1712,1842,2162,2269,2423,2472,2584,2736,2774,2795,2842,2905,3374,3398],"sources":[12,29,40,43,81,121,183,239,265,320,403,448,463,478,532,584,601,774,782,789,791,793,798,845,1082,1087],"evidence":[26,87,88,199,883,1006,1145,1505,1506,1797,2009,2010,2173,2370,2806,2916,3070,3119,3243,3440,3441,3512,3554,3555,3556,3639,3640,3746,3747,4634,4673,4674,4675]},"version":"1lg47t7"}