{"entities":[{"id":"deepmind","name":"DeepMind"},{"id":"google-brain","name":"Google Brain"},{"id":"eliezer-yudkowsky","name":"Eliezer Yudkowsky"},{"id":"rsp-v2-1","name":"Anthropic Responsible Scaling Policy — v2.1"},{"id":"fsf-v3-0","name":"Google DeepMind Frontier Safety Framework — v3.0"},{"id":"preparedness-v2","name":"OpenAI Preparedness Framework — Version 2"},{"id":"guodong-zhang","name":"Guodong Zhang"},{"id":"openai-preferences-announcement","name":"Learning from human preferences — OpenAI account (2017)"},{"id":"meta","name":"Meta"},{"id":"research-supercluster","name":"AI Research SuperCluster (RSC)"},{"id":"qwen-team","name":"Qwen Team / 通义千问"},{"id":"qwen25-announcement","name":"Qwen2.5: A Party of Foundation Models!"},{"id":"gpai-code-of-practice-2025","name":"General-Purpose AI Code of Practice (final version, 10 July 2025)"},{"id":"christiano-2018-minimal-circuits-daemon-free","name":"Open question: are minimal circuits daemon-free?"},{"id":"observed-behaviour-as-evidence","name":"Observed behaviour under training or oversight as evidence"},{"id":"feedback-instils-values","name":"Training on human feedback instils human values"},{"id":"warning-signs-in-time","name":"Clear warning signs will come before danger"},{"id":"extropians-2000-09-05-yudkowsky-4277","name":"Re: Why would AI want to be friendly? (Was: Congratulations to Eli,Brian ...)"},{"id":"lw-2023-alexander-and-yudkowsky-on-agi-goals","name":"Alexander and Yudkowsky on AGI goals"},{"id":"sl4-2001-02-04-yudkowsky-0102-0539","name":"Re: Beyond evolution"},{"id":"sl4-2006-06-06-yudkowsky-0606-15174","name":"Re: [agi] Two draft papers: AI and existential risk; heuristics and biases"},{"id":"automation-lowers-labour-income","name":"Automation lowers the labour share, wages or employment"},{"id":"agrawal-gans-goldfarb-2019-ambiguous-labor-market-impact","name":"Artificial Intelligence: The Ambiguous Labor Market Impact of Automating Prediction"},{"id":"character-ai","name":"Character.AI"},{"id":"karandeep-anand","name":"Karandeep Anand"},{"id":"karol-hausman","name":"Karol Hausman"},{"id":"moonshot-ai","name":"Moonshot AI / 月之暗面"},{"id":"physical-intelligence","name":"Physical Intelligence"},{"id":"yang-zhilin","name":"Yang Zhilin"},{"id":"us-eo-14110-2023","name":"Executive Order 14110: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence"},{"id":"cn-dc-green-low-carbon-plan-2024","name":"Special Action Plan for the Green and Low-Carbon Development of Data Centres (数据中心绿色低碳发展专项行动计划)"},{"id":"cn-ndrc","name":"National Development and Reform Commission (国家发展和改革委员会)"},{"id":"us-hhs-synthetic-dsdna-guidance-2010","name":"Screening Framework Guidance for Providers of Synthetic Double-Stranded DNA"}],"claims":[{"id":"zhang-brain-student-2019","statement":"Guodong Zhang’s CV records a Student Researcher role at Google Brain from 2019-02 to 2019-05.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Retrospective CV; internship/student status is retained. Separate entries are not merged into continuous employment.","evidenceIds":["evidence-zhang-brain-student-2019"],"attributes":{"role":"Student Researcher"}},{"id":"atlas-preferences-credits-deepmind","statement":"OpenAI’s human-preferences account credits collaboration with DeepMind’s safety team.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Specific collaboration credit, not a general influence or personnel-transfer claim.","evidenceIds":["evidence-atlas-preferences-credits-deepmind"]},{"id":"xm-rsc-plan","statement":"Meta planned to expand RSC to 16,000 GPUs during 2022.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A target announced in January 2022, not proof of completion within that year.","evidenceIds":["evidence-xm-rsc-plan"],"attributes":{"transactionType":"compute_capacity","relationshipContext":"internal","transactionStage":"planned","quantity":{"value":16000,"unit":"GPUs","bound":"exact"}}},{"id":"cn-qwen25-release","statement":"Qwen Team published the Qwen2.5 announcement later cited in DeepSeek-R1.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Document publication provides a traceable source path; it does not imply collaboration with its later citers.","evidenceIds":["evidence-cn-qwen25-release"]},{"id":"assessed-preparedness-v2-high-deployment-mitigation","statement":"Preparedness Framework v2’s governance arrangements permit OpenAI Leadership to deploy models with High capabilities without safeguards that sufficiently minimize their associated risks.","derivation":"posed","resolution":"contested","editorialState":"seed","qualification":"Contested assessment of the policy’s permissiveness, not a finding that OpenAI deployed a model without safeguards, a legal ruling or a guarantee of compliance. Coggins et al. (arXiv:2509.24394; v1 2025-09-29, v2 2025-10-13) read the framework’s governance arrangements as letting OpenAI Leadership, the CEO or a person designated by them, approve or reject the Safety Advisory Group’s recommendations and so deploy even more dangerous capabilities, especially if other developers do so. OpenAI’s announcement of Version 2 states that covered systems reaching High capability must have safeguards that sufficiently minimize the associated risk of severe harm before they are deployed, and its ChatGPT agent system card describes the framework as committing it to such safeguards; both predate the analysis and are not replies to it. The record does not decide whether leadership’s authority over the recommendations defeats that stated requirement. The paper’s framework-wide conclusion and OpenAI’s comparison of Version 2 with its predecessor are context. The rule’s own wording is recorded in the provision record for Version 2.","evidenceIds":["evidence-assessed-preparedness-v2-high-deployment-mitigation-coggins-conclusion","evidence-assessed-preparedness-v2-high-deployment-mitigation-coggins-ceo","evidence-assessed-preparedness-v2-high-deployment-mitigation-openai","evidence-assessed-preparedness-v2-high-deployment-mitigation-agent-card","evidence-assessed-preparedness-v2-high-deployment-mitigation-openai-high"],"attributes":{"provision":{"kind":"required_action","key":"pf-high-deployment-restriction"}}},{"id":"prov-rsp-v2-1-may-proceed-with-safeguards","statement":"RSP v2.1 states that if, after its evaluations, Anthropic determines that it has met the ASL-3 Required Safeguards, it may proceed with deploying and training models above the Capability Threshold, provided it has also conducted a follow-up capability assessment.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"What the edition states, quoted from the cited expression. It is not evidence of implementation, compliance or effectiveness.","evidenceIds":["evidence-prov-rsp-v2-1-may-proceed-with-safeguards"],"attributes":{"provision":{"kind":"exception","key":"rsp-may-proceed-with-safeguards"}}},{"id":"prov-fsf-v3-0-response-plan","statement":"FSF v3.0 states that when a model reaches an alert threshold for a CCL, Google DeepMind will assess the model's proximity to the CCL and analyse the risk posed, involving internal and external experts as needed, and that this will inform the formulation and application of a response plan.","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-response-plan"],"attributes":{"provision":{"kind":"required_action","key":"fsf-response-plan"}}},{"id":"ts-b5-christiano-2018-daemons-context","statement":"Christiano’s post of 2018-05-05 says a daemon found by search may predict well only instrumentally and may stop doing so under different conditions.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The same passage as the B3 record.","evidenceIds":["evidence-ts-b5-christiano-2018-daemons-context"],"attributes":{"thought":{"roles":["context"],"register":"nonfiction","missing":"The daemon may stop predicting well under other conditions, but the passage does not state that good predictions observed during the search are weak evidence."}}},{"id":"yk-y05-lw-2023-alexander-and-yudkowsky-on-agi-goals-c","statement":"The post “Alexander and Yudkowsky on AGI goals” (2023-01-24) contests Y05. Answering a question about whether AIs will internalise prohibitions the way most children do, he argues that children's moral learning relies on innate human mechanisms that AI training does not reproduce.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Premise addressed: AI training will instil values the way upbringing instils morals in children. The quotation is a turn attributed to Yudkowsky in a dialogue. Two-author dialogue post (primary author Scott Alexander), described as a lightly edited transcript of a chat held the year before posting (2023-01-24). Only Yudkowsky's turn is quoted. Lightly edited transcript of a conversation held before the posting date; the date shown is the publication date.","evidenceIds":["evidence-yk-t05-d"],"attributes":{"thought":{"roles":["contesting"],"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 AI training will instil values the way upbringing instils morals in children."}}}},{"id":"yk-y08-sl4-2006-06-06-yudkowsky-0606-15174-c","statement":"The SL4 message “Re: [agi] Two draft papers: AI and existential risk; heuristics and biases” (2006-06-06) contests Y08. Replying to Ben Goertzel's call for 'AGI experimentation' as an answer to Friendly AI concerns, he says there is a major context change between the prehuman and posthuman stages, so observed good behaviour does not show that the AI will stay good.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Premise addressed: Dangerous failure modes will show up first in weaker systems, where they can be observed and fixed. Serves as the later rejection for his SL4 message of 2000-07-09.","evidenceIds":["evidence-yk-t08-s5"],"attributes":{"thought":{"roles":["contesting"],"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 dangerous failure modes will show up first in weaker systems, where they can be observed and fixed."}}}},{"id":"prov-eu-code-tr-downstream-14-days","statement":"Measure 1.2 of the Code’s Transparency chapter commits signatories to provide additional information that downstream providers request and that is necessary for them to understand the model’s capabilities and limitations and to meet their obligations, within a reasonable timeframe and no later than 14 days of receiving the request, save for exceptional circumstances.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A commitment that signatories make, subject to the intellectual-property and confidentiality conditions that the measure states. 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-prov-eu-code-tr-downstream-14-days-1","evidence-prov-eu-code-tr-downstream-14-days-2"],"attributes":{"provision":{"kind":"threshold","key":"code-tr-downstream-14-days"}}},{"id":"ec-m06a-agrawal-gans-goldfarb-2019-ambiguous-labor-market-impact-a","statement":"Agrawal, Gans, Goldfarb, “Artificial Intelligence: The Ambiguous Labor Market Impact of Automating Prediction” (2019-02) articulates M06a. The authors argue that workers whose core skill is prediction may lose, while those whose core skill complements prediction may gain.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Hedged ('may'); the paper says the net labour effect cannot yet be assessed.","evidenceIds":["evidence-ec-m06a-20"],"attributes":{"thought":{"roles":["articulation"],"register":"nonfiction","tier":{"value":1,"basis":"NBER working paper (core venue)."}}}},{"id":"lg-deal-anand-character-ai-ceo-2025","statement":"Character.AI announced on 20 June 2025 that Karandeep Anand had joined as Chief Executive Officer.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Company announcement; the post also states he had been a Board Advisor for the previous nine months. It does not establish a start date.","evidenceIds":["lg-ev-deal-anand-character-ai-ceo-2025","lg-ev-deal-anand-character-ai-ceo-2025-advisor"],"attributes":{"role":"Chief Executive Officer"}},{"id":"lg-cn-yang-zhilin-moonshot-founder","statement":"The South China Morning Post reported on 7 August 2024 that Yang Zhilin was the founder of Moonshot AI.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The record rests on reporting, used because no retrievable Moonshot or Yang Zhilin page naming him as founder was found (his personal page says only that he is “working on a startup”). Other co-founders named in reporting are not recorded here.","evidenceIds":["lg-ev-cn-yang-zhilin-moonshot-founder","lg-ev-cn-yang-zhilin-moonshot-founder-startup"]},{"id":"lg-gw-hausman-cofounded-physical-intelligence","statement":"Karol Hausman’s personal homepage, as retrieved on 15 October 2026, describes him as a co-founder of Physical Intelligence (Pi).","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A first-person description on an undated page. It does not establish the founding date, his title or the full founder roster; Physical Intelligence’s own homepage does not name its founders.","evidenceIds":["lg-ev-gw-hausman-cofounded-physical-intelligence"]},{"id":"ap-eliezer-yudkowsky-extropians-2000-09-05-yudkowsky-4277","statement":"The Extropians message “Re: Why would AI want to be friendly? (Was: Congratulations to Eli,Brian ...)” of 2000-09-05, archive message 4277, is sent under the name Eliezer S. Yudkowsky.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A sender name on a mailing-list message as archived; it does not establish identity beyond the credited name, and quoted text by others inside the message is not the sender’s.","evidenceIds":["evidence-ap-eliezer-yudkowsky-extropians-2000-09-05-yudkowsky-4277"]},{"id":"ap-eliezer-yudkowsky-sl4-2001-02-04-yudkowsky-0102-0539","statement":"The SL4 message “Re: Beyond evolution” of 2001-02-04, archive message 0539, 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-sl4-2001-02-04-yudkowsky-0102-0539"]},{"id":"cmp-fsf-v3-0-d2-levels","statement":"FSF · v3.0 states a provision on the comparison question “Named levels or thresholds”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Misuse: CBRN uplift level 1, Cyber uplift level 1, Harmful manipulation level 1 (exploratory). ML R&D: ML R&D acceleration level 1, ML R&D automation level 1. Misalignment (illustrative): Instrumental Reasoning Levels 1 and 2. Early warning evaluations carry “alert thresholds” meant to signal that a CCL could be reached ahead of the next risk assessment (PDF p. 5). What the edition states, paraphrased from the cited expression with its page. It is not evidence of implementation, compliance or effectiveness, and a shared question does not make two editions’ provisions equivalent.","evidenceIds":["ev-cmp-fsf-v3-0-d2-levels-1"],"attributes":{"provision":{"kind":"threshold","key":"comparison-d2-levels","facet":{"id":"D2.levels","status":"stated"}}}},{"id":"cpo-pw-cn-dc-plan-issued","statement":"The NDRC, MIIT, NEA and National Data Administration issued the Special Action Plan for the Green and Low-Carbon Development of Data Centres under document number 发改环资〔2024〕970号, dated 3 July 2024.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Records issuance only.","evidenceIds":["evidence-cpo-pw-cn-dc-plan-issued-1","evidence-cpo-pw-cn-dc-plan-issued-2","evidence-cpo-pw-cn-dc-plan-issued-3"]},{"id":"ab-scr-hhs-2010-voluntary-all-providers","statement":"The 2010 Guidance states that following it is voluntary, though many of its specific recommendations serve to remind providers of their obligations under existing regulations.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The notice's statement about its own standing, not a finding about any provider's practice; the existing regulations it refers to are not recorded here. Split under the one-claim-per-edge rule from the record that also carried the Guidance's addressees, which are now the separate record recorded separately. The provision kind is the nearest the dataset schema admits (the schema has no kind for an instrument's standing): the sentence states that the Guidance itself imposes no duty beyond obligations existing regulations already carry. See openQuestions. The provision kind is now `standing` (schema 2.22), the kind for the standing a text claims for itself, where part 1 used `exception`.","evidenceIds":["evidence-ab-scr-hhs-2010-voluntary-all-providers-1"],"attributes":{"provision":{"kind":"standing","key":"us-hhs-screening-guidance-2010-voluntary-standing"}}},{"id":"ab-gov-us-eo14110-sec44-purpose","statement":"Section 4.4 of Executive Order 14110, headed \"Reducing Risks at the Intersection of AI and CBRN Threats\", states that the actions it directs are to better understand and mitigate the risk of AI being misused to assist in the development or use of CBRN threats, with a particular focus on biological weapons.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"States the order's own purpose for this section. It is not evidence that any action was taken under it, and it names no AI system. Executive Order 14110 was revoked by Executive Order 14148 of 20 January 2025 (recorded separately); the revocation does not erase the text recorded here.","evidenceIds":["evidence-ab-gov-us-eo14110-sec44-purpose-1"],"attributes":{"provision":{"kind":"risk_domain","key":"us-eo14110-sec44a-ai-cbrn-purpose"}}}],"sources":[{"id":"source-pf-v2-update","title":"Our updated Preparedness Framework","authors":["OpenAI"],"publisher":"OpenAI","publicationDate":"2025-04-15","version":"Dated Version 2 publication announcement","canonicalUrl":"https://openai.com/index/updating-our-preparedness-framework/","evidenceUrl":"https://openai.com/index/updating-our-preparedness-framework/","language":"en","archivedUrl":"https://web.archive.org/web/20250416173732/https://openai.com/index/updating-our-preparedness-framework/","archivedAt":"2025-04-16","accessedAt":"2026-09-18","reuse":{"attribution":"OpenAI, Updating our Preparedness Framework.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-rsp-2-1","title":"Responsible Scaling Policy — v2.1","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Responsible Scaling Policy v2.1; effective date 2025-03-31.","versionDate":"2025-03-31","canonicalUrl":"https://www-cdn.anthropic.com/17310f6d70ae5627f55313ed067afc1a762a4068.pdf","evidenceUrl":"https://www-cdn.anthropic.com/17310f6d70ae5627f55313ed067afc1a762a4068.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250402134435/https://www-cdn.anthropic.com/17310f6d70ae5627f55313ed067afc1a762a4068.pdf","archivedAt":"2025-04-02","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v2.1.","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-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-zhang-cv","title":"Guodong Zhang — curriculum vitae","authors":["Guodong Zhang"],"publisher":"Guodong Zhang","publicationDate":null,"version":"Undated four-page CV inspected 2026-09-19; PDF p. 1 Professional Experience visually inspected. Publication date not inferred from entries, metadata or retrieval. University of Toronto-hosted CV.","canonicalUrl":"https://www.cs.toronto.edu/~gdzhang/docs/resume.pdf","evidenceUrl":"https://www.cs.toronto.edu/~gdzhang/docs/resume.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20230713103323/http://www.cs.toronto.edu/~gdzhang/docs/resume.pdf","archivedAt":"2023-07-13","accessedAt":"2026-09-19","reuse":{"attribution":"Guodong Zhang, Guodong Zhang — curriculum vitae","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-preferences","title":"Learning from human preferences","authors":["Dario Amodei","Paul Christiano","Alex Ray"],"publisher":"OpenAI","publicationDate":"2017-06-13","version":"Dated publisher page inspected 2026-09-28; current HTML, not an authenticated historical capture.","canonicalUrl":"https://openai.com/index/learning-from-human-preferences/","evidenceUrl":"https://openai.com/index/learning-from-human-preferences/","language":"en","archivedUrl":"https://web.archive.org/web/20260926013813/https://openai.com/index/learning-from-human-preferences/","archivedAt":"2026-09-26","accessedAt":"2026-09-28","reuse":{"attribution":"Dario Amodei; Paul Christiano; Alex Ray. Learning from human preferences.","licence":"No redistribution licence asserted.","note":"Short attributed paraphrases only; original publication is linked, not redistributed."}},{"id":"source-xm-rsc-2022","title":"Introducing the AI Research SuperCluster — Meta’s cutting-edge AI supercomputer for AI research","authors":["Kevin Lee","Shubho Sengupta"],"publisher":"Meta","publicationDate":"2022-01-24","version":"January 24, 2022 announcement; January 28 privacy update explicitly distinguished.","canonicalUrl":"https://ai.meta.com/blog/ai-rsc/","evidenceUrl":"https://ai.meta.com/blog/ai-rsc/","language":"en","archivedUrl":"https://web.archive.org/web/20260921080617/https://ai.meta.com/blog/ai-rsc/","archivedAt":"2026-09-21","accessedAt":"2026-09-29","reuse":{"attribution":"Introducing the AI Research SuperCluster — Meta’s cutting-edge AI supercomputer for AI research — Meta","licence":"No redistribution licence asserted.","note":"Source-linked metadata and brief evidence paraphrases; original text remains with its publisher."}},{"id":"source-cn-qwen25","title":"Qwen2.5: A Party of Foundation Models!","authors":["Qwen Team"],"publisher":"Qwen Team","publicationDate":"2024-09-19","version":"Source language: English (en). Dated team announcement; mutable page, inspected at retrieval.","canonicalUrl":"https://qwenlm.github.io/blog/qwen2.5/","evidenceUrl":"https://qwenlm.github.io/blog/qwen2.5/","language":"en","archivedUrl":"https://web.archive.org/web/20261003124153/https://qwenlm.github.io/blog/qwen2.5/","archivedAt":"2026-10-03","accessedAt":"2026-09-29","reuse":{"attribution":"Qwen Team, Qwen2.5: A Party of Foundation Models! (2024-09-19).","licence":"No redistribution licence asserted for this source expression","note":"Bibliographic metadata and brief attributed paraphrases only; model licensing is distinct from source-document reuse."}},{"id":"source-coggins-pf-v2","title":"The 2025 OpenAI Preparedness Framework does not guarantee any AI risk mitigation practices: a proof-of-concept for affordance analyses of AI safety policies","authors":["Sam Coggins","Alexander K. Saeri","Katherine A. Daniell","Lorenn P. Ruster","Jessie Liu","Jenny L. Davis"],"publisher":"arXiv","publicationDate":"2025-09-29","version":"arXiv:2509.24394v2, a preprint: v1 submitted 2025-09-29, v2 revised 2025-10-13. The abstract is quoted from the v2 page; the archived copy is the Internet Archive capture of the unversioned abstract page on 2025-12-05, after v2 was posted.","versionDate":"2025-10-13","canonicalUrl":"https://arxiv.org/abs/2509.24394","evidenceUrl":"https://arxiv.org/abs/2509.24394v2","language":"en","archivedUrl":"https://web.archive.org/web/20251205152906/https://arxiv.org/abs/2509.24394","archivedAt":"2025-12-05","accessedAt":"2026-10-03","reuse":{"attribution":"Sam Coggins et al., The 2025 OpenAI Preparedness Framework does not guarantee any AI risk mitigation practices, arXiv:2509.24394v2 (2025).","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-openai-chatgpt-agent-system-card","title":"ChatGPT Agent System Card","authors":["OpenAI"],"publisher":"OpenAI","publicationDate":"2025-07-17","version":"System card dated 17 July 2025 on its cover. The quoted sentence and its page match the PDF served on 2026-10-04 and the Internet Archive capture of 2025-07-17; the two copies differ only in chart labels.","canonicalUrl":"https://openai.com/index/chatgpt-agent-system-card/","evidenceUrl":"https://cdn.openai.com/pdf/839e66fc-602c-48bf-81d3-b21eacc3459d/chatgpt_agent_system_card.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250717225133/https://cdn.openai.com/pdf/839e66fc-602c-48bf-81d3-b21eacc3459d/chatgpt_agent_system_card.pdf","archivedAt":"2025-07-17","accessedAt":"2026-10-04","reuse":{"attribution":"OpenAI, ChatGPT Agent System Card (2025-07-17).","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-christiano-2018-minimal-circuits-daemon-free","title":"Open question: are minimal circuits daemon-free?","authors":["Paul Christiano"],"publisher":"Alignment Forum and LessWrong","publicationDate":"2018-05-05","version":"Text checked in the post’s first stored version through the LessWrong GraphQL API on 2026-10-06; content version 1.0.0, never edited.","canonicalUrl":"https://www.alignmentforum.org/posts/nyCHnY7T5PHPLjxmN/open-question-are-minimal-circuits-daemon-free","evidenceUrl":"https://www.alignmentforum.org/posts/nyCHnY7T5PHPLjxmN/open-question-are-minimal-circuits-daemon-free","language":"en","archivedUrl":"https://web.archive.org/web/20260624174850/https://www.alignmentforum.org/posts/nyCHnY7T5PHPLjxmN/open-question-are-minimal-circuits-daemon-free","archivedAt":"2026-06-24","accessedAt":"2026-10-06","reuse":{"attribution":"Paul Christiano, \"Open question: are minimal circuits daemon-free?\", Alignment Forum and LessWrong, 2018-05-05.","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-extropians-2000-09-05-yudkowsky-4277","title":"Re: Why would AI want to be friendly? (Was: Congratulations to Eli,Brian ...) (Extropians mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"Extropians mailing list","publicationDate":"2000-09-05","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 817049e5c2b33859a370b003534043a70de9c896f3dd32f7fbb9dfe948097f0b); 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.3Q00/4277.html","evidenceUrl":"http://extropians.weidai.com/extropians.3Q00/4277.html","language":"en","archivedUrl":"https://web.archive.org/web/20160414051453/http://extropians.weidai.com/extropians.3Q00/4277.html","archivedAt":"2016-04-14","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Re: Why would AI want to be friendly? (Was: Congratulations to Eli,Brian ...)”, Extropians mailing list, 2000-09-05.","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-lw-2023-alexander-and-yudkowsky-on-agi-goals","title":"Alexander and Yudkowsky on AGI goals","authors":["Scott Alexander","Eliezer Yudkowsky"],"publisher":"LessWrong","publicationDate":"2023-01-24","version":"Post body retrieved through the LessWrong GraphQL API on 2026-10-06 (SHA-256 3be0140f09cb16ea35ba143237ecda5eb86458f7168ff0a6bab0dd6237e345f3), content version 1.0.0.","canonicalUrl":"https://www.lesswrong.com/posts/rwkkcgSpnAyE8oNo3/alexander-and-yudkowsky-on-agi-goals","evidenceUrl":"https://www.lesswrong.com/posts/rwkkcgSpnAyE8oNo3/alexander-and-yudkowsky-on-agi-goals","language":"en","archivedUrl":"https://web.archive.org/web/20260910002502/https://www.lesswrong.com/posts/rwkkcgSpnAyE8oNo3/alexander-and-yudkowsky-on-agi-goals","archivedAt":"2026-09-10","accessedAt":"2026-10-06","reuse":{"attribution":"Scott Alexander and Eliezer Yudkowsky, “Alexander and Yudkowsky on AGI goals”, LessWrong, 2023-01-24.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-sl4-2001-02-04-yudkowsky-0102-0539","title":"Re: Beyond evolution (SL4 mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"SL4 mailing list","publicationDate":"2001-02-04","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 52b0e09dc587e4dfb66c3a32c19f110f86f750abc59596229bffd6dd0960ba06); 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/0102/0539.html","evidenceUrl":"http://sl4.org/archive/0102/0539.html","language":"en","archivedUrl":"https://web.archive.org/web/20240425142601/http://sl4.org/archive/0102/0539.html","archivedAt":"2024-04-25","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Re: Beyond evolution”, SL4 mailing list, 2001-02-04.","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":"source-sl4-2006-06-06-yudkowsky-0606-15174","title":"Re: [agi] Two draft papers: AI and existential risk; heuristics and biases (SL4 mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"SL4 mailing list","publicationDate":"2006-06-06","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 c2e944580a586039db6294e2d5b2c4fffaf9aefcaf50965480ecd12788934b58); 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/0606/15174.html","evidenceUrl":"http://sl4.org/archive/0606/15174.html","language":"en","archivedUrl":"https://web.archive.org/web/20260309154322/http://sl4.org/archive/0606/15174.html","archivedAt":"2026-03-09","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Re: [agi] Two draft papers: AI and existential risk; heuristics and biases”, SL4 mailing list, 2006-06-06.","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-eu-code-transparency-chapter","title":"General-Purpose AI Code of Practice: Transparency Chapter","authors":["Nuria Oliver (Working Group 1 Co-Chair)","Rishi Bommasani (Working Group 1 Vice-Chair)"],"publisher":"European Commission","publicationDate":"2025-07-10","version":"PDF retrieved 2026-10-11 from the Commission download address (SHA-1 d3b14e82fc376661ec8edbad5e57dc814b74fcd3). The Code page gives 10 July 2025 as the publication date of the Code; the PDF carries no date. 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/118120","evidenceUrl":"https://ec.europa.eu/newsroom/dae/redirection/document/118120","language":"en","archivedUrl":"https://web.archive.org/web/20260913173946/https://ec.europa.eu/newsroom/dae/redirection/document/118120","archivedAt":"2026-09-13","accessedAt":"2026-10-11","reuse":{"attribution":"European Commission, General-Purpose AI Code of Practice, Transparency Chapter (10 July 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-agrawal-gans-goldfarb-2019-ambiguous-labor-market-impact","title":"Artificial Intelligence: The Ambiguous Labor Market Impact of Automating Prediction","authors":["Ajay Agrawal","Joshua S. Gans","Avi Goldfarb"],"publisher":"NBER Working Paper 25619","publicationDate":"2019-02","version":"Full text. Current copy retrieved directly on 2026-10-09; no archive capture (Wayback unreachable), so earlier bytes are not authenticated. SHA-256 dbd015967a47f77a41ff8028f5cc8208c3bb0e8cb44d6b1cc9be04f7df9c2450. Journal version not retrieved. Agrawal, A., Gans, J. S., & Goldfarb, A. (2019). Artificial intelligence: The ambiguous labor market impact of automating prediction. Journal of Economic Perspectives, 33(2), 31–50. https://doi.org/10.1257/jep.33.2.31","canonicalUrl":"https://www.nber.org/system/files/working_papers/w25619/w25619.pdf","evidenceUrl":"https://www.nber.org/system/files/working_papers/w25619/w25619.pdf","language":"en","accessedAt":"2026-10-09","reuse":{"attribution":"Ajay Agrawal, Joshua S. Gans, Avi Goldfarb, “Artificial Intelligence: The Ambiguous Labor Market Impact of Automating Prediction”, NBER Working Paper 25619, 2019-02.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"lg-src-cn-scmp-yang-2024","title":"Moonshot AI founder builds business in the mould of ByteDance, OpenAI","authors":["Ben Jiang"],"publisher":"South China Morning Post","publicationDate":"2024-08-07","version":"Dated page, retrieved on 2026-10-15 (SHA-256 4712dc8c82136f63e81f90cb029b6247ba9b96599d335a51fd0479a621090ad9) through the third-party reader relay r.jina.ai, which renders the publisher’s current page (the publisher refused direct retrieval); not an archive copy.","canonicalUrl":"https://www.scmp.com/tech/big-tech/article/3273461/meet-yang-zhilin-moonshot-ai-founder-builds-business-mould-bytedance-openai","evidenceUrl":"https://www.scmp.com/tech/big-tech/article/3273461/meet-yang-zhilin-moonshot-ai-founder-builds-business-mould-bytedance-openai","language":"en","archivedUrl":"https://web.archive.org/web/20260805103644/https://www.scmp.com/tech/big-tech/article/3273461/meet-yang-zhilin-moonshot-ai-founder-builds-business-mould-bytedance-openai","archivedAt":"2026-08-05","accessedAt":"2026-10-15","reuse":{"attribution":"Ben Jiang, “Moonshot AI founder builds business in the mould of ByteDance, OpenAI”, 2024-08-07.","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":"lg-src-cn-yang-zhilin-home","title":"Zhilin Yang (personal website)","authors":["Zhilin Yang"],"publisher":"Zhilin Yang","publicationDate":null,"version":"Undated post, retrieved on 2026-10-15 (SHA-256 c265a1dd5d6fac0b966f90e0c7d50bf0f6061535e321d4b1636489094ec2b623) directly from the publisher. Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://kimiyoung.github.io/","evidenceUrl":"https://kimiyoung.github.io/","language":"en","archivedUrl":"https://web.archive.org/web/20260906051756/https://kimiyoung.github.io/","archivedAt":"2026-09-06","accessedAt":"2026-10-15","reuse":{"attribution":"Zhilin Yang, “Zhilin Yang (personal website)”.","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-cai-anand-2025","title":"Character.AI Names Karandeep Anand as CEO","authors":["Character.AI"],"publisher":"Character.AI","publicationDate":"2025-06-20","version":"Dated page, retrieved on 2026-10-15 (SHA-256 1bea9ff4537b8fab14daf82024ada97a3cec74e46805d740ef8a97468c277219) directly from the publisher.","canonicalUrl":"https://blog.character.ai/character-ai-names-karandeep-anand-as-ceo/","evidenceUrl":"https://blog.character.ai/character-ai-names-karandeep-anand-as-ceo/","language":"en","archivedUrl":"https://web.archive.org/web/20261015210623/https://blog.character.ai/character-ai-names-karandeep-anand-as-ceo/","archivedAt":"2026-10-15","accessedAt":"2026-10-15","reuse":{"attribution":"Character.AI, “Character.AI Names Karandeep Anand as CEO”, 2025-06-20.","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-gw-hausman-home","title":"Karol Hausman (personal homepage, undated)","authors":["Karol Hausman"],"publisher":"Karol Hausman","publicationDate":null,"version":"Undated post, retrieved on 2026-10-15 (SHA-256 dc2a3dcaf4733ef53e2381f1b486f43ac038e61788ebbe1112e2eba03fe030aa) directly from the publisher. Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://karolhausman.github.io/","evidenceUrl":"https://karolhausman.github.io/","language":"en","archivedUrl":"https://web.archive.org/web/20260629133322/https://karolhausman.github.io/","archivedAt":"2026-06-29","accessedAt":"2026-10-15","reuse":{"attribution":"Karol Hausman, “Karol Hausman (personal homepage, undated)”.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"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-cpo-pw-cn-dc-green-plan-2024","title":"关于印发《数据中心绿色低碳发展专项行动计划》的通知 (发改环资〔2024〕970号)","authors":["国家发展改革委 工业和信息化部 国家能源局 国家数据局"],"publisher":"中国政府网 (gov.cn)","publicationDate":"2024-07-03","version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 8160516ee0bfeed37c38601d295670624b640084d2a3768c91c2c1c04a6139d6). gov.cn State Council policy library page as served on 2026-10-28. Retrieved again at admission on 2026-10-30 (SHA-256 8160516ee0bfeed37c38601d295670624b640084d2a3768c91c2c1c04a6139d6, the same bytes).","canonicalUrl":"https://www.gov.cn/zhengce/zhengceku/202407/content_6964165.htm","evidenceUrl":"https://www.gov.cn/zhengce/zhengceku/202407/content_6964165.htm","language":"zh-Hans","accessedAt":"2026-10-28","reuse":{"attribution":"国家发展改革委 工业和信息化部 国家能源局 国家数据局, “关于印发《数据中心绿色低碳发展专项行动计划》的通知 (发改环资〔2024〕970号)”, 2024-07-03.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-ab-hhs-2010-fr","title":"Screening Framework Guidance for Providers of Synthetic Double-Stranded DNA (75 FR 62820)","authors":["Federal Register (Office of the Federal Register, National Archives) via govinfo.gov"],"publisher":"Federal Register (Office of the Federal Register, National Archives) via govinfo.gov","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 d3bfd95a1e3ae4d212505081166f0f8bb7475845e225ae643e6d4a03a93b7e56). Federal Register print of 13 October 2010, Vol. 75, No. 197, pp. 62820-62832, FR Doc. 2010-25728. The notice publishes the final Guidance in full, after the comment responses. Retrieved again at admission on 2026-10-30 (SHA-256 d3bfd95a1e3ae4d212505081166f0f8bb7475845e225ae643e6d4a03a93b7e56, the same bytes).","canonicalUrl":"https://www.govinfo.gov/content/pkg/FR-2010-10-13/pdf/2010-25728.pdf","evidenceUrl":"https://www.govinfo.gov/content/pkg/FR-2010-10-13/pdf/2010-25728.pdf","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"Federal Register (Office of the Federal Register, National Archives) via govinfo.gov, “Screening Framework Guidance for Providers of Synthetic Double-Stranded DNA (75 FR 62820)”.","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-zhang-brain-student-2019","sourceRole":"primary","support":"supports","locator":"PDF p. 1, Professional Experience, Student Researcher / Google Brain row","passage":"The CV dates this Student Researcher entry 2019-02 to 2019-05.","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-atlas-preferences-credits-deepmind","sourceRole":"primary","support":"supports","locator":"Opening paragraph and final paragraph before Footnote","passage":"OpenAI explicitly describes the research as collaboration with DeepMind’s safety team.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-28","note":"Original source passage checked by the machine agent that proposed the record. No independent machine audit or human source inspection is recorded for this increment."},{"id":"evidence-xm-rsc-plan","sourceRole":"primary","support":"supports","locator":"“Phase two and beyond”, paragraph on growing from 6,080 to 16,000 GPUs.","passage":"Meta planned to expand RSC to 16,000 GPUs during 2022.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-29","note":"Passage checked by machine against the identified source expression. No human editorial acceptance or independent verification of implementation is implied."},{"id":"evidence-cn-qwen25-release","sourceRole":"primary","support":"supports","locator":"Page title, September 19, 2024 dateline and Qwen Team byline","passage":"Qwen Team announces the Qwen2.5 family.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-29","note":"English source paraphrased and checked by machine. Exploratory seed, not editorially accepted. Source reporting is not independent verification of implementation."},{"id":"evidence-assessed-preparedness-v2-high-deployment-mitigation-coggins-conclusion","sourceRole":"secondary","support":"context","locator":"PDF p. 14, Conclusions, first sentence (arXiv:2509.24394v2; identical in v1)","passage":"We conclude that OpenAI’s April 2025 Preparedness Framework does not guarantee any AI risk mitigation practices.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-03 against the v2 PDF (arXiv:2509.24394v2, revised 2025-10-13); identical in v1 (2025-09-29), and the paper’s title states the same conclusion. Recorded as context on 2026-10-04: the conclusion is framework-wide and rests also on evaluation scope and Medium capabilities, not only on deployment governance. Human editorial review is pending."},{"id":"evidence-assessed-preparedness-v2-high-deployment-mitigation-coggins-ceo","sourceRole":"secondary","support":"supports","locator":"PDF p. 11, Results, heading “Key result #3” and the following paragraph (arXiv:2509.24394v2; identical in v1)","passage":"Key result #3) The Preparedness Framework allows the OpenAI CEO to deploy even more dangerous capabilities, especially if other developers do so.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-03 against the v2 PDF (arXiv:2509.24394v2, revised 2025-10-13); identical in v1 (2025-09-29). The following paragraph grounds it in the framework’s provisions that OpenAI Leadership can approve or reject the Safety Advisory Group’s recommendations and can unilaterally bypass them. The authors’ reading of the framework’s governance arrangements: evidence of the disagreement, not of what the framework says or how it is applied. Human editorial review is pending."},{"id":"evidence-assessed-preparedness-v2-high-deployment-mitigation-openai","sourceRole":"primary","support":"context","locator":"Opening, second paragraph, first sentence","passage":"stronger requirements for what it means to “sufficiently minimize” those risks in practice","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt (mid-sentence) checked by an automated research assistant on 2026-10-03 against the live page and the Internet Archive capture of 2025-04-16. OpenAI’s characterisation of its own update, published with the edition. Reclassified as context on 2026-10-04: it compares Version 2 with its predecessor (stronger requirements than before) and is compatible with the critics’ position; it does not say the rule binds OpenAI. Human editorial review is pending."},{"id":"evidence-prov-rsp-v2-1-may-proceed-with-safeguards","sourceRole":"primary","support":"supports","locator":"PDF p. 14 (printed p. 10), §4.3 Safeguards Decision, first paragraph, first sentence","passage":"If, after the evaluations above, we determine that we have met the ASL-3 Required Safeguards, then we may proceed with deploying and training models above the Capability Threshold, provided we have also conducted a follow-up capability assessment.","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 2025-04-02. Extraction note: Wording identical to v2.0; no change proposed. Human editorial review is pending."},{"id":"evidence-prov-fsf-v3-0-response-plan","sourceRole":"primary","support":"supports","locator":"PDF p. 6 (printed p. 6), §1.4 Response Plans and Mitigations, second paragraph, first two sentences","passage":"When a model reaches an alert threshold for a CCL, we will assess the proximity of the model to the CCL and analyze the risk posed, involving internal and external experts as needed. This will inform the formulation and application of a response plan.","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: Wording identical to v2.0 (PDF p. 3, component 3, first paragraph); no change proposed. Human editorial review is pending."},{"id":"evidence-assessed-preparedness-v2-high-deployment-mitigation-agent-card","sourceRole":"primary","support":"contradicts","locator":"PDF p. 16 (printed p. 15), §5 Preparedness Framework, first paragraph, second sentence","passage":"The framework commits us to track and mitigate the risk of severe harm, including by implementing safeguards that sufficiently minimize the risk for highly capable models.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-04","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-04 against the PDF served by OpenAI and the Internet Archive capture of 2025-07-17 (same page and wording). OpenAI presents the framework as committing it to implement safeguards that sufficiently minimize the risk, corroborating the publisher position stated in the announcement; not independent proof of a guarantee or of effect. It predates Coggins et al. and is not a reply to them; the same section reports that OpenAI applied the rule to this launch, which this record does not assess. Human editorial review is pending."},{"id":"evidence-assessed-preparedness-v2-high-deployment-mitigation-openai-high","sourceRole":"primary","support":"contradicts","locator":"Bullet “Clarified capability levels”, sentence beginning “Covered systems that reach High capability”","passage":"Covered systems that reach High capability must have safeguards that sufficiently minimize the associated risk of severe harm before they are deployed.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-04","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-04 against the live page and the Internet Archive capture of 2025-04-16. OpenAI’s stated predeployment requirement, scoped to covered systems that reach High capability: the publisher’s position, not independent evidence of compliance or of whether leadership’s authority can override it. Published with the edition, before the analysis; not a reply to it. Human editorial review is pending."},{"id":"evidence-ts-b5-christiano-2018-daemons-context","sourceRole":"primary","support":"supports","locator":"Third opening prose paragraph, beginning \"I may get a daemon\", after the introductory note","passage":"a consequentialist who happens to be motivated to make good predictions (perhaps because it has realized that only good predictors survive). Under different conditions, the daemon may no longer be motivated to predict well","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-06","note":"The same passage as the B3 record."},{"id":"evidence-yk-t05-d","sourceRole":"primary","support":"supports","locator":"Alexander and Yudkowsky on AGI goals, section 1, [Yudkowsky][13:36]","passage":"If AI is built on anything like the present paradigm, or on future paradigms either really, you can't map that onto the complicated particular mechanisms that get invoked by raising a human child, and expect the same result.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-06","note":"Candidate pack passage T05-d. The quotation (38 words) matched the retrieved expression (SHA-256 3be0140f09cb16ea35ba143237ecda5eb86458f7168ff0a6bab0dd6237e345f3) when checked on 2026-10-06."},{"id":"evidence-yk-t08-s5","sourceRole":"primary","support":"supports","locator":"message body, paragraph beginning 'First, as discussed in the chapter'","passage":"I can think of *many* failure modes such that the AI appears to behave well as a prehuman, then wipes out humanity as a posthuman.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-07","note":"Candidate pack passage T08-s5. The quotation (25 words) matched the retrieved expression (SHA-256 c2e944580a586039db6294e2d5b2c4fffaf9aefcaf50965480ecd12788934b58) when checked on 2026-10-07."},{"id":"evidence-prov-eu-code-tr-downstream-14-days-1","sourceRole":"primary","support":"supports","locator":"Commitment 1, Measure 1.2, third paragraph","passage":"Signatories will provide such information within a reasonable timeframe, and no later than 14 days of receiving the request save for exceptional circumstances.","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-prov-eu-code-tr-downstream-14-days-2","sourceRole":"primary","support":"supports","locator":"Commitment 1, Measure 1.2, third paragraph, preceding sentence","passage":"\"Such information\" is additional information requested by downstream providers insofar as it is necessary to enable them to have a good understanding of the capabilities and limitations of the model relevant for its integration and to comply with their obligations under the AI Act.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-11","note":"Editorial paraphrase proposed by the proposing machine agent from the text 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-ec-m06a-20","sourceRole":"primary","support":"supports","locator":"pdf page 6; introduction (printed p. 4)","passage":"On the other hand, workers whose core skill is prediction, such as human resource workers who screen resumes, may find the value of their occupation diminished.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-14","note":"Candidate pack passage M06a-20 (research/economics-mainstream). The quotation (26 words) matched the retrieved current copy (SHA-256 dbd015967a47f77a41ff8028f5cc8208c3bb0e8cb44d6b1cc9be04f7df9c2450) in the pack's verification. Re-checked at admission on 2026-10-14 against a copy retrieved that day.","contextOnly":{"reason":"Fresh-context audit (A3) judged that the quotation misleads when shown alone.","requires":"Same paragraph: authors 'cannot assess the net effect of artificial intelligence on labor, even in the short run'; the opposite effect for non-prediction workers is stated in the preceding sentence."}},{"id":"lg-ev-deal-anand-character-ai-ceo-2025","sourceRole":"primary","support":"supports","locator":"Paragraph 1","passage":"Karandeep Anand has joined Character.AI as Chief Executive Officer.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (9 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read directly from the publisher. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. No human editorial review.","retrievalRoute":"direct"},{"id":"lg-ev-deal-anand-character-ai-ceo-2025-advisor","sourceRole":"primary","support":"context","locator":"Paragraph 2","passage":"he’s spent the past nine months as a Board Advisor","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. No human editorial review.","retrievalRoute":"direct"},{"id":"lg-ev-cn-yang-zhilin-moonshot-founder","sourceRole":"secondary","support":"supports","locator":"Opening paragraph (text available before the subscription barrier)","passage":"The report calls Yang Zhilin the founder and chief executive of Moonshot AI.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The paraphrase’s verbatim anchor “founder and chief executive Yang Zhilin” 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. 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":"reader_relay"},{"id":"lg-ev-cn-yang-zhilin-moonshot-founder-startup","sourceRole":"primary","support":"context","locator":"About section, first sentence","passage":"I am working on a startup.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (6 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read 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-gw-hausman-cofounded-physical-intelligence","sourceRole":"primary","support":"supports","locator":"Introduction, first sentence","passage":"I'm a co-founder of Physical Intelligence (Pi)","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (7 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read directly from the publisher. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ap-eliezer-yudkowsky-extropians-2000-09-05-yudkowsky-4277","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-sl4-2001-02-04-yudkowsky-0102-0539","sourceRole":"primary","support":"supports","locator":"Message header, sender name (address not reproduced)","passage":"Eliezer S. Yudkowsky","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-27","note":"Credited name checked by an automated research assistant on 2026-10-27 in a direct retrieval and checked by a reviewer from a different model family on 2026-10-27. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"ev-cmp-fsf-v3-0-d2-levels-1","sourceRole":"primary","support":"supports","locator":"PDF p. 4 (printed p. 4), §1.2 Critical Capability Levels, second paragraph","passage":"The edition describes three sets of CCLs: misuse, machine learning R&D and misalignment.","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":"evidence-cpo-pw-cn-dc-plan-issued-1","sourceRole":"primary","support":"supports","locator":"metadata block","passage":"发文字号：发改环资〔2024〕970号","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, compute, power and orbit (research/compute-power-orbit). The quotation (10 words, each CJK character counted as one) 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. English machine translation, not verified by a person: “Document number: Fagai Huanzi [2024] No. 970”. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-cpo-pw-cn-dc-plan-issued-2","sourceRole":"primary","support":"supports","locator":"metadata block","passage":"成文日期：2024年07月03日","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, compute, power and orbit (research/compute-power-orbit). The quotation (10 words, each CJK character counted as one) 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. English machine translation, not verified by a person: “Date of writing: 3 July 2024”. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-cpo-pw-cn-dc-plan-issued-3","sourceRole":"primary","support":"supports","locator":"metadata block","passage":"发文机关：国家发展改革委 工业和信息化部 国家能源局 国家数据局","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, compute, power and orbit (research/compute-power-orbit). The quotation (28 words, each CJK character counted as one) 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. English machine translation, not verified by a person: “Issuing bodies: NDRC, MIIT, National Energy Administration, National Data Administration”. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-scr-hhs-2010-voluntary-all-providers-1","sourceRole":"primary","support":"supports","locator":"SUMMARY, 75 FR 62820","passage":"Following this Guidance is voluntary, though many specific recommendations serve to remind providers of their obligations under existing regulations.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (19 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-gov-us-eo14110-sec44-purpose-1","sourceRole":"primary","support":"supports","locator":"Sec. 4.4(a), 88 FR 75200","passage":"To better understand and mitigate the risk of AI being misused to assist in the development or use of CBRN threats—with a particular focus on biological weapons","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (27 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. The section heading reads \"Reducing Risks at the Intersection of AI and CBRN Threats\". No human editorial review.","retrievalRoute":"direct"}],"at":{"entities":[9,10,19,43,51,54,78,127,157,170,180,184,253,328,345,354,357,389,438,456,486,518,547,644,716,719,747,768,829,1038,1079,1107,1215],"claims":[184,287,350,373,452,533,549,957,1022,1046,1228,1408,1768,1953,1986,2381,2394,2617,3043,3285,3326],"sources":[33,37,46,84,127,156,166,188,190,333,356,405,423,453,465,499,585,586,590,674,874,939,1044],"evidence":[203,308,373,396,570,571,572,599,616,684,685,1180,1249,1274,1525,1526,1754,2274,2275,2522,2523,2565,3028,3041,3285,3957,3958,3959,4502,4570]},"version":"1lg47t7"}