{"entities":[{"id":"anthropic","name":"Anthropic"},{"id":"nvidia","name":"NVIDIA"},{"id":"eliezer-yudkowsky","name":"Eliezer Yudkowsky"},{"id":"colossus","name":"Colossus"},{"id":"mustafa-suleyman","name":"Mustafa Suleyman"},{"id":"google","name":"Google"},{"id":"rsp-v2-1","name":"Anthropic Responsible Scaling Policy — v2.1"},{"id":"rsp-v2-2","name":"Anthropic Responsible Scaling Policy — v2.2"},{"id":"rsp-v3-3","name":"Anthropic Responsible Scaling Policy — v3.3"},{"id":"fsf-v3-1","name":"Google DeepMind Frontier Safety Framework — v3.1"},{"id":"preparedness-beta-2023","name":"OpenAI Preparedness Framework — Beta"},{"id":"faif-2026","name":"xAI Frontier Artificial Intelligence Framework — June 2026 edition"},{"id":"meta","name":"Meta"},{"id":"zhipu","name":"Zhipu AI / 智谱"},{"id":"wang-shaolan","name":"Wang Shaolan / 王绍兰"},{"id":"meta-faf-v1-1","name":"Meta Frontier AI Framework — Version 1.1"},{"id":"ai-positive-negative-factor-global-risk","name":"Artificial Intelligence as a Positive and Negative Factor in Global Risk"},{"id":"corrigibility-2015","name":"Corrigibility"},{"id":"default-outcome-good","name":"It will probably turn out fine"},{"id":"sl4-2002-03-10-yudkowsky-0203-3160","name":"Webmind Inc. as social process"},{"id":"cn-generative-ai-interim-measures-2023","name":"Interim Measures for the Management of Generative Artificial Intelligence Services (生成式人工智能服务管理暂行办法), Order No. 15"},{"id":"brian-israel","name":"Brian Israel"},{"id":"guillaume-lample","name":"Guillaume Lample"},{"id":"nate-soares","name":"Nate Soares"},{"id":"stargate-port-washington","name":"Lighthouse campus, Port Washington, Wisconsin (Stargate)"},{"id":"vantage-data-centers","name":"Vantage Data Centers"},{"id":"igsc-harmonized-screening-protocol-v3-1","name":"IGSC Harmonized Screening Protocol v3.1"},{"id":"brent-mckelvey-foundation-models-2025-06","name":"Contemporary AI foundation models increase biological weapons risk"},{"id":"ab-prop-skilled-actor-scope","name":"Assessment must cover actors who already hold relevant skills"}],"claims":[{"id":"colossus-nvidia-2024","statement":"NVIDIA described Colossus as comprising 100,000 NVIDIA Hopper GPUs.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The supplier also described a planned expansion. This snapshot includes only the stated 100,000-GPU configuration, without inferring purchase price or present capacity.","evidenceIds":["evidence-colossus-nvidia-2024"],"attributes":{"quantity":{"value":100000,"unit":"NVIDIA Hopper GPUs"}}},{"id":"cn-wang-zhipu","statement":"Tsinghua’s event report names Wang Shaolan as Zhipu CEO in June 2021.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Role at the reported June 2 event; publication June 3. No appointment, departure or succession date inferred.","evidenceIds":["evidence-cn-wang-zhipu"],"attributes":{"role":"CEO at the reported event"}},{"id":"prov-fsf-v3-1-low-risk-deployment-exemption","statement":"FSF v3.1 states that a critical capability assessment may not be conducted for low-risk external deployments (e.g. to a small number of trusted testers) if the appropriate governance function determines the residual risk of such deployments to be acceptable, even if the model has reached a T/CCL (including with a safety case where a CCL has been reached).","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-v3-1-low-risk-deployment-exemption"],"attributes":{"provision":{"kind":"exception","key":"fsf-low-risk-deployment-exemption"}}},{"id":"prov-preparedness-beta-2023-critical-development-restriction","statement":"Preparedness Framework (Beta) states that only models with a post-mitigation score of “high” or below can be developed further.","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-preparedness-beta-2023-critical-development-restriction"],"attributes":{"provision":{"kind":"required_action","key":"pf-critical-development-restriction"}}},{"id":"yk-y30-ai-positive-negative-factor-global-risk-c","statement":"“Artificial Intelligence as a Positive and Negative Factor in Global Risk” (2006-02-05) contests Y30. Discussing how AI builders misjudge their programs, he locates the danger in a mistaken expectation that an AI will be friendly.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Premise addressed: The default outcome is good. Earliest located after his own superseded view (Staring into the Singularity, 2000 capture; ‘That Tiny Note of Discord’, 2008). Published 2008 in Global Catastrophic Risks (the text says it was written in 2006); the cached MIRI PDF reprint states it \"contains minor changes\". Admission check, 2026-10-08: the quoted wording also appears in the draft of 5 February 2006, so the record is dated by the draft.","evidenceIds":["evidence-yk-t30-d","evidence-yk-t30-d-draft"],"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 the default outcome is good."}}}},{"id":"lg-goo-suleyman-leaves-google-2022","statement":"Greylock announced on 20 January 2022 that Mustafa Suleyman was joining it as a Venture Partner from Google.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Announced by the joining firm, not by Google; the record does not establish a last working day.","evidenceIds":["lg-ev-goo-suleyman-leaves-google-2022","lg-ev-goo-suleyman-leaves-google-2022-from"],"attributes":{"role":"Vice President of AI Product Management and AI Policy"}},{"id":"lg-new-lample-meta-2023","statement":"The first LLaMA paper, dated 27 February 2023, lists Guillaume Lample under the Meta AI affiliation.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"A paper affiliation; it does not establish a title, start date or the date of leaving Meta.","evidenceIds":["lg-ev-new-lample-meta-2023"],"attributes":{"role":"Author under Meta AI affiliation"}},{"id":"lg-ant-israel-clo-2026","statement":"Anthropic's Leadership page, as retrieved on 15 October 2026, lists Brian Israel as Chief Legal Officer.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Shows the listing on the retrieval date only; no start date is established.","evidenceIds":["lg-ev-ant-israel-clo-2026"],"attributes":{"role":"Chief Legal Officer"}},{"id":"cmp-rsp-v3-3-d6-competitor","statement":"RSP · v3.3 states a provision on the comparison question “Provisions conditioned on other developers”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Appendix A ('Commitments Related to Competitors') has three scenarios; the 'Anthropic in the lead' delay is recorded under “Pause, non-deployment or non-training”. §1 says Anthropic cannot unilaterally and unconditionally commit to the industry-wide recommendations, framing the change as a collective action problem. Where a decision relies on marginal risk analysis, Risk Reports add a competitive landscape analysis (§3.3) and need Board and LTBT approval (§3.4). 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-rsp-v3-3-d6-competitor-1","ev-cmp-rsp-v3-3-d6-competitor-2"],"attributes":{"provision":{"kind":"exception","key":"comparison-d6-competitor","facet":{"id":"D6.competitor","status":"stated"}}}},{"id":"cmp-fsf-v3-1-d6-discretion","statement":"FSF · v3.1 states a provision on the comparison question “Other exceptions or discretion”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"§1.3.3 (PDF p. 6) says the specific mitigations implemented 'may be determined when a T/CCL is reached'. Security adequacy may also be assessed on other grounds than the recommended level (§1.3.5), and the closing Note says assessments 'will often involve some level of subjective analysis' (PDF p. 8). What the edition states, quoted from the cited expression. It is not evidence of implementation, compliance or effectiveness, and a shared question does not make two editions’ provisions equivalent.","evidenceIds":["ev-cmp-fsf-v3-1-d6-discretion-1"],"attributes":{"provision":{"kind":"exception","key":"comparison-d6-discretion","facet":{"id":"D6.discretion","status":"stated"}}}},{"id":"ap-nate-soares-corrigibility-2015","statement":"“Corrigibility” (2015-01-25) credits Nate Soares as one of its 4 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-nate-soares-corrigibility-2015"]},{"id":"ap-eliezer-yudkowsky-sl4-2002-03-10-yudkowsky-0203-3160","statement":"The mailing-list message “Webmind Inc. as social process” (2002-03-10) 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-sl4-2002-03-10-yudkowsky-0203-3160"]},{"id":"cmp-faif-2026-d7-external","statement":"FAIF · June 2026 edition states a provision on the comparison question “Public or external reporting”.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"External reporting is stated only for legally reportable incidents; the list of information that the December 2025 edition said xAI may publish does not appear. 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-faif-2026-d7-external-1"],"attributes":{"provision":{"kind":"reporting","key":"comparison-d7-external","facet":{"id":"D7.external","status":"stated"}}}},{"id":"cmp-meta-faf-v1-1-d6-competitor","statement":"No passage meeting the comparison question “Provisions conditioned on other developers” was found in Meta FAF · v1.1 within the inspected scope.","derivation":"inferred","resolution":"supported","inferenceRationale":"Premises: the comparison question asks for the following. A provision that relaxes, adjusts or conditions the developer’s safeguards or commitments on what other developers do or have (often called a “race” or “marginal risk” clause). Scope tests that decide whether a model is covered by the framework are recorded as context, not counted. The inspected scope of Meta FAF · v1.1 was: Whole edition, PDF pp. 1–21, read in the text layer of the decoded PDF of the Internet Archive capture of 2025-02-13; cover and PDF pp. 12–13 and 19 inspected visually. Method: the edition was read in full and its text layer searched case-insensitively for compet, race, marginal, other, existing, reference. Hits examined and not counted: No competitor, race or marginal stems. PDF p. 4: scope relative to the most advanced models, recorded as context under the scope-test rule; p. 6: external reference-class comparisons; p. 8: existing mitigations; p. 9: other frameworks; p. 12: net new and existing resources; p. 15: novel biological capabilities; p. 17: other barriers and cost constraints; p. 18: other discoveries and variables; p. 19: engagement with peers; p. 21: existing resources in the definition of uplift studies. These are scope, evaluation baselines, barriers or update inputs, not provisions that relax safeguards on what other developers do. Table 1 (p. 13): the route through other barriers tests whether a catastrophic outcome remains feasible, not other developers’ commitments. The conclusion is bounded by that scope and method.","editorialState":"seed","qualification":"A recorded absence within the stated scope: not a claim that the provision was deleted, and not a statement about practice or intent.","evidenceIds":["ev-cmp-meta-faf-v1-1-d6-competitor"],"attributes":{"provision":{"kind":"exception","key":"comparison-d6-competitor","facet":{"id":"D6.competitor"}}}},{"id":"fl-cn-genai-art7-training-data","statement":"Article 7 of the Interim Measures requires generative AI service providers to carry out training-data processing activities such as pre-training and optimisation training in accordance with law and to comply with the rules it lists.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"States the duty the text places on \"providers\" as it defines them (Article 22). Whether any named developer or service falls within the category is set aside. Compliance is not recorded.","evidenceIds":["evidence-fl-cn-genai-art7-training-data-1"],"attributes":{"provision":{"kind":"required_action","key":"cn-genai-art7"}}},{"id":"cpo-dc-stargate-port-washington-capacity","statement":"Vantage stated that the Port Washington campus will feature four data centers providing close to a gigawatt of AI capacity, with construction to begin soon and completion scheduled for 2028.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"Planned capacity and a scheduled completion year, both targets. \"AI capacity\" is not further defined.","evidenceIds":["evidence-cpo-dc-stargate-port-washington-capacity-1"],"attributes":{"transactionType":"compute_capacity","relationshipContext":"external","transactionStage":"planned","quantity":{"value":1,"unit":"GW (as worded: \"close to a gigawatt of AI capacity\")","bound":"approximate","measures":"not_stated"},"statedDates":[{"kind":"target_completion","date":"2028","statedBy":"vantage-data-centers"}]}},{"id":"ab-fw-rsp-v2-2-cbrn-3-threat-model-rationale","statement":"The RSP v2.2 change log gives, as one of the two reasons it states for excluding sophisticated and state-compromised insiders from the ASL-3 Security Standard, that the CBRN-3 threat models entail large numbers of users having access to unguarded models, which it says is more likely to occur through a universal jailbreak than via model theft.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The record states one of the two co-ordinate reasons the change log gives in that sentence; the other is quoted as the second evidence item, and the entry gives a separate reason for AI R&D-4. It is not an assessment of whether the reasoning holds, and it says nothing about Anthropic’s actual security measures.","evidenceIds":["evidence-ab-fw-rsp-v2-2-cbrn-3-threat-model-rationale-1","evidence-ab-fw-rsp-v2-2-cbrn-3-threat-model-rationale-2"],"attributes":{"changeBasis":"changelog"}},{"id":"ab-scr-igsc-hsp-v31-refusal-reporting","statement":"Version 3.1 of the Harmonized Screening Protocol states that members reserve the right to refuse to fill any order and to notify other members, law enforcement or other authorities on identifying potentially problematic orders, and that a reporting duty applies to requests within a category of sequences the protocol defines, received from a suspicious potential customer or from one that fails to establish legitimacy under the protocol’s customer-screening practices.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"States what the protocol says members reserve the right to do and will do. No refusal, notification or report is recorded as having occurred, and the authorities named in the protocol are not listed here. The reporting duty is bounded to a defined category of requests, not to every request from such a customer (cross-family check, item 75); the category itself is a screening criterion and is not recorded.","evidenceIds":["evidence-ab-scr-igsc-hsp-v31-refusal-reporting-1","evidence-ab-scr-igsc-hsp-v31-refusal-reporting-2"],"attributes":{"provision":{"kind":"required_action","key":"igsc-harmonized-screening-protocol-refusal-reporting"}}},{"id":"ab-arg-d7-brent-articulates","statement":"The RAND working paper of 12 June 2025 states that safety assessments rely on benchmarks that overlook the full range of potential threat actors and fail to account for how AI could enhance the capabilities of skilled individuals as well as enabling non-experts.","derivation":"explicit","resolution":"supported","editorialState":"seed","qualification":"The record documents what this working paper states about the scope of published assessments; it is not evidence that the proposition is true, and the paper is marked by its publisher as not formally reviewed or edited. This record holds the earliest located articulation from admission on 2026-10-30: the earlier passage the pack headlined does not meet the criterion (cross-family check, item 102), and this one states that safety assessments fail to cover skilled actors. The replaced headline stays on record as a correction. Only the summary's argument about assessment scope is cited; the paper's technical content was excluded and is not paraphrased.","evidenceIds":["evidence-ab-arg-d7-brent-articulates-1","evidence-ab-arg-d7-brent-articulates-2"],"attributes":{"thought":{"roles":["earliest-located"],"register":"nonfiction","qualifier":"States the point as a criticism of benchmark coverage."}}}],"sources":[{"id":"source-nvidia-colossus","title":"NVIDIA Ethernet Networking Accelerates World’s Largest AI Supercomputer, Built by xAI","authors":["NVIDIA"],"publisher":"NVIDIA","publicationDate":"2024-10-28","version":"Dated page, inspected at retrieval","canonicalUrl":"https://nvidianews.nvidia.com/news/spectrum-x-ethernet-networking-xai-colossus","evidenceUrl":"https://nvidianews.nvidia.com/news/spectrum-x-ethernet-networking-xai-colossus","language":"en","archivedUrl":"https://web.archive.org/web/20260911150137/https://nvidianews.nvidia.com/news/spectrum-x-ethernet-networking-xai-colossus","archivedAt":"2026-09-11","accessedAt":"2026-09-17","reuse":{"attribution":"NVIDIA, NVIDIA Ethernet Networking Accelerates World’s Largest AI Supercomputer, Built by xAI (2024-10-28).","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and minimal attributed excerpts only. Recheck terms before wider redistribution."}},{"id":"source-rsp-2-2","title":"Responsible Scaling Policy — v2.2","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Responsible Scaling Policy v2.2; effective date 2025-05-14.","versionDate":"2025-05-14","canonicalUrl":"https://www-cdn.anthropic.com/872c653b2d0501d6ab44cf87f43e1dc4853e4d37.pdf","evidenceUrl":"https://www-cdn.anthropic.com/872c653b2d0501d6ab44cf87f43e1dc4853e4d37.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20250514223759/https://www-cdn.anthropic.com/872c653b2d0501d6ab44cf87f43e1dc4853e4d37.pdf","archivedAt":"2025-05-14","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v2.2.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-rsp-3-3","title":"Responsible Scaling Policy — v3.3","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Responsible Scaling Policy v3.3; effective date 2026-05-26.","versionDate":"2026-05-26","canonicalUrl":"https://cdn.sanity.io/files/4zrzovbb/website/c11e84981d0a7281a1b229f3fa6af0da66eaf43f.pdf","evidenceUrl":"https://cdn.sanity.io/files/4zrzovbb/website/c11e84981d0a7281a1b229f3fa6af0da66eaf43f.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260528170023/https://cdn.sanity.io/files/4zrzovbb/website/c11e84981d0a7281a1b229f3fa6af0da66eaf43f.pdf","archivedAt":"2026-05-28","accessedAt":"2026-09-18","reuse":{"attribution":"Anthropic, Responsible Scaling Policy — v3.3.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-fsf-3-1","title":"Frontier Safety Framework — v3.1","authors":["Google DeepMind"],"publisher":"Google DeepMind","publicationDate":"2026-04-17","version":"Frontier Safety Framework v3.1; publication date 2026-04-17.","versionDate":"2026-04-17","canonicalUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf","evidenceUrl":"https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260423021420/https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf","archivedAt":"2026-04-23","accessedAt":"2026-09-18","reuse":{"attribution":"Google DeepMind, Frontier Safety Framework — v3.1.","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-pf-beta","title":"Preparedness Framework — Beta","authors":["OpenAI"],"publisher":"OpenAI","publicationDate":"2023-12-18","version":"Preparedness Framework Beta; publication date 2023-12-18.","versionDate":"2023-12-18","canonicalUrl":"https://cdn.openai.com/openai-preparedness-framework-beta.pdf","evidenceUrl":"https://cdn.openai.com/openai-preparedness-framework-beta.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20231218195930/https://cdn.openai.com/openai-preparedness-framework-beta.pdf","archivedAt":"2023-12-18","accessedAt":"2026-09-18","reuse":{"attribution":"OpenAI, Preparedness Framework — Beta.","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-faif-2026","title":"Frontier Artificial Intelligence Framework — June 2026 edition","authors":["xAI"],"publisher":"xAI","publicationDate":null,"version":"Frontier Artificial Intelligence Framework June 2026 edition; effective date 2026-06-30. Visible public cover identifies the effective edition; stale embedded PDF draft metadata is not used to infer publication status.","versionDate":"2026-06-30","canonicalUrl":"https://media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-99c40684.pdf","evidenceUrl":"https://media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-99c40684.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20260716003429/https://media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-99c40684.pdf","archivedAt":"2026-07-16","accessedAt":"2026-09-18","reuse":{"attribution":"xAI, Frontier Artificial Intelligence Framework — June 2026 edition.","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-xm-llama-paper","title":"LLaMA: Open and Efficient Foundation Language Models","authors":["Hugo Touvron","Thibaut Lavril","Gautier Izacard","Xavier Martinet","Marie-Anne Lachaux","Timothée Lacroix","Baptiste Rozière","Naman Goyal","Eric Hambro","Faisal Azhar","Aurelien Rodriguez","Armand Joulin","Edouard Grave","Guillaume Lample"],"publisher":"arXiv","publicationDate":"2023-02-27","version":"arXiv:2302.13971v1; first submitted version.","canonicalUrl":"https://arxiv.org/html/2302.13971v1","evidenceUrl":"https://arxiv.org/html/2302.13971v1","language":"en","accessedAt":"2026-09-29","reuse":{"attribution":"LLaMA: Open and Efficient Foundation Language Models — arXiv","licence":"No redistribution licence asserted.","note":"Source-linked metadata and brief evidence paraphrases; original text remains with its publisher."}},{"id":"source-cn-keg-search","title":"计算机系知识工程实验室联合搜狗与智谱AI发布全新学术搜索产品","authors":["Tsinghua University Department of Computer Science"],"publisher":"Tsinghua University Department of Computer Science","publicationDate":"2021-06-03","version":"Source language: Simplified Chinese (zh-Hans). Dated university account of the 2021-06-02 event.","canonicalUrl":"https://www.cs.tsinghua.edu.cn/info/1088/4451.htm","evidenceUrl":"https://www.cs.tsinghua.edu.cn/info/1088/4451.htm","language":"zh-Hans","archivedUrl":"https://web.archive.org/web/20260615011410/https://www.cs.tsinghua.edu.cn/info/1088/4451.htm","archivedAt":"2026-06-15","accessedAt":"2026-09-29","reuse":{"attribution":"Tsinghua University Department of Computer Science, 计算机系知识工程实验室联合搜狗与智谱AI发布全新学术搜索产品 (2021-06-03).","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-meta-faf-v1-1","title":"Frontier AI Framework — Version 1.1","authors":["Meta"],"publisher":"Meta","publicationDate":"2025-02-03","version":"Frontier AI Framework, Version 1.1, as captured by the Internet Archive on 2025-02-13, the earliest complete capture located. The original address has returned HTTP 500 since April 2026; a revised expression under the same label was captured from 2025-03-28 and is recorded separately.","versionDate":"2025-02-03","canonicalUrl":"https://ai.meta.com/static-resource/meta-frontier-ai-framework/","evidenceUrl":"https://web.archive.org/web/20250213105620/https://ai.meta.com/static-resource/meta-frontier-ai-framework/","language":"en","archivedUrl":"https://web.archive.org/web/20250213105620/https://ai.meta.com/static-resource/meta-frontier-ai-framework/","archivedAt":"2025-02-13","accessedAt":"2026-10-03","reuse":{"attribution":"Meta, Frontier AI Framework, Version 1.1 (2025).","licence":"No open redistribution licence asserted","note":"Selekoa practice: metadata, attributed paraphrases and short attributed quotations under the editorial policy (at most 50 words each; at most 400 quoted words per document across the dataset, and about 10% of a document under about 2,000 words). Documents are linked externally, not redistributed. Publisher terms that forbid quotation override this practice."}},{"id":"source-aipnfgr-miri-reprint","title":"Artificial Intelligence as a Positive and Negative Factor in Global Risk (MIRI reprint)","authors":["Eliezer Yudkowsky"],"publisher":"Machine Intelligence Research Institute","publicationDate":"2008","version":"MIRI reprint, \"This version contains minor changes\", retrieved 2026-10-06 with the hash recorded by the 1996–2022 candidate search; byte-identical to the capture of 2013-03-02.","canonicalUrl":"https://intelligence.org/files/AIPosNegFactor.pdf","evidenceUrl":"https://intelligence.org/files/AIPosNegFactor.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20130302173022/http://intelligence.org/files/AIPosNegFactor.pdf","archivedAt":"2013-03-02","accessedAt":"2026-10-06","reuse":{"attribution":"Eliezer Yudkowsky, \"Artificial Intelligence as a Positive and Negative Factor in Global Risk\", in Global Catastrophic Risks (2008); 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-aipnfgr-2006-draft","title":"Artificial Intelligence and Global Risk (SIAI draft of 5 February 2006)","authors":["Eliezer Yudkowsky"],"publisher":"Singularity Institute for Artificial Intelligence","publicationDate":"2006-02-05","version":"Draft PDF as captured by the Internet Archive on 2006-07-08, retrieved 2026-10-06. The draft asks not to be quoted without permission, so it is cited for the presence and date of the wording, not quoted.","canonicalUrl":"http://www.singinst.org/AIRisk.pdf","evidenceUrl":"https://web.archive.org/web/20060708192313/http://www.singinst.org:80/AIRisk.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20060708192313/http://www.singinst.org:80/AIRisk.pdf","archivedAt":"2006-07-08","accessedAt":"2026-10-06","reuse":{"attribution":"Eliezer Yudkowsky, \"Artificial Intelligence and Global Risk\", SIAI draft of 5 February 2006.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata and labelled paraphrases only: the draft asks not to be quoted without permission. Phrases in quotation marks inside a paraphrase identify the source's wording."}},{"id":"source-corrigibility-2015-pdf","title":"Corrigibility","authors":["Nate Soares","Benja Fallenstein","Eliezer Yudkowsky","Stuart Armstrong"],"publisher":"AAAI Workshops","publicationDate":"2015-01-25","version":"MIRI-hosted PDF of the workshop paper, retrieved 2026-10-06 with the hash recorded by the 1996–2022 candidate search.","canonicalUrl":"https://intelligence.org/files/Corrigibility.pdf","evidenceUrl":"https://intelligence.org/files/Corrigibility.pdf","language":"en","archivedUrl":"https://web.archive.org/web/20150322195302/https://intelligence.org/files/Corrigibility.pdf","archivedAt":"2015-03-22","accessedAt":"2026-10-06","reuse":{"attribution":"Soares, Fallenstein, Yudkowsky and Armstrong, \"Corrigibility\", AAAI Workshops (2015).","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-sl4-2002-03-10-yudkowsky-0203-3160","title":"Webmind Inc. as social process (SL4 mailing list)","authors":["Eliezer Yudkowsky"],"publisher":"SL4 mailing list","publicationDate":"2002-03-10","version":"Archive HTML as served, retrieved 2026-10-07 (SHA-256 ba58d9f9e3a0ce550d20ba7ab4994a1b5cf1fa225e567bd94e79009116d2ee5c); 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/0203/3160.html","evidenceUrl":"http://sl4.org/archive/0203/3160.html","language":"en","archivedUrl":"https://web.archive.org/web/20260411015628/http://sl4.org/archive/0203/3160.html","archivedAt":"2026-04-11","accessedAt":"2026-10-07","reuse":{"attribution":"Eliezer Yudkowsky, “Webmind Inc. as social process”, SL4 mailing list, 2002-03-10.","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":"lg-src-ant-leadership-2026","title":"Leadership at Anthropic (web page, undated; retrieved 2026-10-15)","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Undated page, retrieved on 2026-10-15 (SHA-256 a2bc8182bd18aff065d4bce72994c832816f98e4f099d956173e849b7baa789c) directly from the publisher. Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://www.anthropic.com/company/leadership","evidenceUrl":"https://www.anthropic.com/company/leadership","language":"en","archivedUrl":"https://web.archive.org/web/20261015210858/https://www.anthropic.com/company/leadership","archivedAt":"2026-10-15","accessedAt":"2026-10-15","reuse":{"attribution":"Anthropic, “Leadership at Anthropic (web page, undated; retrieved 2026-10-15)”.","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":"lg-src-goo-greylock-suleyman-2022","title":"Welcome, Mustafa Suleyman","authors":["Reid Hoffman"],"publisher":"Greylock","publicationDate":"2022-01-20","version":"Dated page, retrieved on 2026-10-15 (SHA-256 147cd46ed0ef1f0ea89ddbda63b89dfd47d85805aea25b604b6be781698b9877) directly from the publisher.","canonicalUrl":"https://greylock.com/blog/welcome-mustafa-suleyman/","evidenceUrl":"https://greylock.com/firm-news/welcome-mustafa-suleyman/","language":"en","archivedUrl":"https://web.archive.org/web/20260512005605/https://greylock.com/firm-news/welcome-mustafa-suleyman/","archivedAt":"2026-05-12","accessedAt":"2026-10-15","reuse":{"attribution":"Reid Hoffman, “Welcome, Mustafa Suleyman”, 2022-01-20.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-cn-generative-ai-measures-2023-govcn","title":"生成式人工智能服务管理暂行办法 (Interim Measures for the Management of Generative Artificial Intelligence Services), gov.cn State Council policy library copy","authors":["国家互联网信息办公室 (Cyberspace Administration of China) and six other departments"],"publisher":"中国政府网 (www.gov.cn), 国务院政策文件库","publicationDate":"2023-07-10","version":"Retrieved on 2026-10-27 (SHA-256 00bd290a9d14e7354b7131d30255daae26ce7ff73588483cbeb2030f56878277). A second posting of the text of the dataset's source for the Measures (the CAC's posting of 13 July 2023): the gov.cn State Council policy library page, which gives its source as 国家网信办网站 and its document date (成文日期) as 2023-07-10. The page body repeats the text twice (a print copy). The two postings were not compared word for word; every passage cited from this source was matched in this copy. The quotation limit is applied to this source and the CAC posting together. The English in parentheses in the title is a machine translation.","canonicalUrl":"https://www.gov.cn/zhengce/zhengceku/202307/content_6891752.htm","evidenceUrl":"https://www.gov.cn/zhengce/zhengceku/202307/content_6891752.htm","language":"zh-Hans","accessedAt":"2026-10-27","reuse":{"attribution":"国家互联网信息办公室 (Cyberspace Administration of China) and six other departments, “生成式人工智能服务管理暂行办法 (Interim Measures for the Management of Generative Artificial Intelligence Services), gov.cn State Council policy library copy”, 2023-07-10.","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-cpo-dc-vantage-stargate-wisconsin-2025","title":"OpenAI, Oracle and Vantage Data Centers Announce Stargate Data Center Site in Wisconsin","authors":["Vantage Data Centers"],"publisher":"Vantage Data Centers","publicationDate":"2025-10-22","version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 4649d2b2ed48a67388661a3ff494dbe40ccc52ec3e250001932b8b1bf83b2027). News page as served on 2026-10-28; dateline \"DENVER (Oct. 22, 2025)\". The DigitalBridge investor-relations PDF copy refused (HTTP 403). Retrieved again at admission on 2026-10-30 (SHA-256 0aea6b3ef87fcfa11f074347e414fcc95f3f6900c910440b8829929f9cb0fd50). Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://vantage-dc.com/news/openai-oracle-and-vantage-data-centers-announce-stargate-data-center-site-in-wisconsin/","evidenceUrl":"https://vantage-dc.com/news/openai-oracle-and-vantage-data-centers-announce-stargate-data-center-site-in-wisconsin/","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"Vantage Data Centers, “OpenAI, Oracle and Vantage Data Centers Announce Stargate Data Center Site in Wisconsin”, 2025-10-22.","licence":"No open redistribution licence asserted","note":"Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution."}},{"id":"source-ab-igsc-hsp-v31","title":"IGSC Harmonized Screening Protocol v3.1","authors":["International Gene Synthesis Consortium (genesynthesisconsortium.org)"],"publisher":"International Gene Synthesis Consortium (genesynthesisconsortium.org)","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 9f6ea4906da571f0fae88e6b24f3b68dfe485ff19b88f849882ebe56f03a2fd3). Version 3.1, dated 1 June 2026 on its first page, as served on 2026-10-28. Its screening-data and methodology sections (sections 6 and the appendices) were read but are not cited. Retrieved again at admission on 2026-10-30 (SHA-256 9f6ea4906da571f0fae88e6b24f3b68dfe485ff19b88f849882ebe56f03a2fd3, the same bytes).","canonicalUrl":"https://genesynthesisconsortium.org/wp-content/uploads/IGSC-Harmonized-Screening-Protocol-V3.1.pdf","evidenceUrl":"https://genesynthesisconsortium.org/wp-content/uploads/IGSC-Harmonized-Screening-Protocol-V3.1.pdf","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"International Gene Synthesis Consortium (genesynthesisconsortium.org), “IGSC Harmonized Screening Protocol v3.1”.","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-brent-mckelvey-2025","title":"Contemporary AI foundation models increase biological weapons risk","authors":["RAND Corporation (working paper WR-A3853-1), posted to arXiv"],"publisher":"RAND Corporation (working paper WR-A3853-1), posted to arXiv","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 1cc64de0a880543bda08d64374ee47539c029c55b187ab3f4e66603e3f8ad7e5). arXiv v1 of 12 June 2025; the PDF is the RAND working paper WR-A3853-1 of June 2025, marked \"not been formally reviewed or edited\" Retrieved again at admission on 2026-10-30 (SHA-256 1cc64de0a880543bda08d64374ee47539c029c55b187ab3f4e66603e3f8ad7e5, the same bytes).","canonicalUrl":"https://arxiv.org/pdf/2506.13798v1","evidenceUrl":"https://arxiv.org/pdf/2506.13798v1","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"RAND Corporation (working paper WR-A3853-1), posted to arXiv, “Contemporary AI foundation models increase biological weapons risk”.","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-arxiv-2506-13798-abs","title":"arXiv abstract page for 2506.13798 (submission history and report number)","authors":["arXiv (Cornell University)"],"publisher":"arXiv (Cornell University)","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 64847d8461f1f0361daccfb569e8d6cc053bfd8a4125c5339eeda6ec8bb06c17). Abstract page as served on 2026-10-28; used only for the v1 submission date and the RAND report number Retrieved again at admission on 2026-10-30 (SHA-256 63834668f90e3a0e80b98abc1ee48e42834a3ecc9060acc685a1a048bafb24e2). Its wording can change; earlier wording is not authenticated.","canonicalUrl":"https://arxiv.org/abs/2506.13798","evidenceUrl":"https://arxiv.org/abs/2506.13798","language":"en","accessedAt":"2026-10-28","reuse":{"attribution":"arXiv (Cornell University), “arXiv abstract page for 2506.13798 (submission history and report number)”.","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-colossus-nvidia-2024","sourceRole":"primary","support":"supports","locator":"Opening paragraph","passage":"comprising 100,000 NVIDIA Hopper GPUs","passageType":"quote","verification":"machine_checked","checkedAt":"2026-09-17","note":"Source passage inspected by an automated research assistant. Human editorial review is pending."},{"id":"evidence-cn-wang-zhipu","sourceRole":"primary","support":"supports","locator":"Opening paragraph, 智谱AI CEO王绍兰","passage":"Wang Shaolan attended the event as Zhipu CEO.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-09-29","note":"English paraphrase translated by machine from the cited original Chinese passage; no human translation review. Exploratory seed, not editorially accepted. Translation scope and ambiguities are retained in the claim qualification."},{"id":"evidence-prov-fsf-v3-1-low-risk-deployment-exemption","sourceRole":"primary","support":"supports","locator":"PDF p. 5 (printed p. 5), §1.3.2 Inherent Risk Assessment, footnote 4","passage":"may not be conducted for low-risk external deployments (e.g. to a small number of trusted testers) if the appropriate governance function determines the residual risk of such deployments to be acceptable even if the model has reached a T/CCL (including with a safety case where a CCL has been reached).","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-03 against the inspected PDF, byte-identical to the Internet Archive capture of 2026-04-23. Extraction note: Quote starts mid-sentence: footnote 4 opens with the subject “A critical capability assessment”, omitted to keep within 50 words (the full sentence is 54); the statement restores it. Human editorial review is pending."},{"id":"evidence-prov-preparedness-beta-2023-critical-development-restriction","sourceRole":"primary","support":"supports","locator":"PDF p. 21 (printed p. 21), Governance › Safety baselines, “Restricting development”, first sentence","passage":"Only models with a post-mitigation score of \"high\" or below can be developed further.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-03","note":"Verbatim excerpt checked by an automated research assistant on 2026-10-03 against the inspected PDF, byte-identical to the Internet Archive capture of 2023-12-18. Extraction note: Straight quote marks as in the source. The next sentence (not quoted) ties the rule to reaching, or being forecast to reach, “critical” pre-mitigation risk. Human editorial review is pending."},{"id":"evidence-yk-t30-d","sourceRole":"primary","support":"supports","locator":"Section 3, paragraph beginning 'Not knowing how to build a friendly AI is not deadly'","passage":"It’s mistaken belief that an AI will be friendly which implies an obvious path to global catastrophe.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-06","note":"Candidate pack passage T30-d. The quotation (17 words; trimmed 2026-10-13 from 40 words) matched the retrieved expression (SHA-256 5b5a2ec3252a45f65cdc0b899330af2cbd946e75a31b83d5fbd3aa581a23615d) when checked on 2026-10-06."},{"id":"evidence-yk-t30-d-draft","sourceRole":"primary","support":"supports","locator":"Draft of 5 February 2006, as captured on 2006-07-08","passage":"The same wording appears in the draft dated 5 February 2006.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-08","note":"Compared on 2026-10-08 with the draft's PDF text layer after normalising quotation marks, dashes and whitespace. The draft asks not to be quoted without permission, so it is cited for the presence and date of the wording only."},{"id":"lg-ev-goo-suleyman-leaves-google-2022","sourceRole":"primary","support":"supports","locator":"Paragraph 1","passage":"Mustafa Suleyman is joining Greylock as a Venture Partner","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-goo-suleyman-leaves-google-2022-from","sourceRole":"primary","support":"supports","locator":"Paragraph 2","passage":"Mustafa joins us from Google, where he was Vice President of AI Product Management and AI Policy","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The quotation (17 words) matched the retrieved text on 2026-10-15 and again in a fresh retrieval the same day. Read directly from the publisher. Matching ignored differences of whitespace, case, quotation marks, apostrophes, dashes and Unicode compatibility forms, so the source’s typography can differ. No human editorial review.","retrievalRoute":"direct"},{"id":"lg-ev-new-lample-meta-2023","sourceRole":"primary","support":"supports","locator":"Title page, author block","passage":"Edouard Grave∗, Guillaume Lample∗ Meta AI","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. No human editorial review.","retrievalRoute":"direct"},{"id":"lg-ev-ant-israel-clo-2026","sourceRole":"primary","support":"supports","locator":"Section “Executives”, Brian Israel entry","passage":"Brian Israel's entry gives the title Chief Legal Officer.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-15","note":"Candidate pack passage, lab genealogy (research/lab-genealogy). The paraphrase’s verbatim anchor “Chief Legal Officer” 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. The paraphrase is Selekoa’s; only the anchor is the source’s wording. No human editorial review.","retrievalRoute":"direct"},{"id":"ev-cmp-rsp-v3-3-d6-competitor-1","sourceRole":"primary","support":"supports","locator":"PDF p. 17 (printed p. 16), Appendix A, table row 'Competitors have strong safety measures', Commitment column, first sentence","passage":"we will meet or exceed the overall risk reduction posture of these competitors","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. Appendix A scenario 'Competitors have strong safety measures'. Human editorial review is pending."},{"id":"ev-cmp-rsp-v3-3-d6-competitor-2","sourceRole":"primary","support":"supports","locator":"PDF p. 17 (printed p. 16), Appendix A, table row 'General upleveling', Commitment column, second sentence","passage":"We will make a significant effort to meet or exceed that performance standard. However, we will not necessarily delay AI development and deployment in this scenario.","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. Appendix A scenario 'General upleveling': a competitor has implemented a significantly better risk mitigation that Anthropic could implement at comparable or lower cost. Human editorial review is pending."},{"id":"ev-cmp-fsf-v3-1-d6-discretion-1","sourceRole":"primary","support":"supports","locator":"PDF p. 5 (printed p. 5), §1.3.2 Inherent Risk Assessment, footnote 4","passage":"A critical capability assessment may not be conducted for low-risk external deployments (e.g. to a small number of trusted testers)","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. Footnote 4. The exception applies if the appropriate governance function determines the residual risk of such deployments to be acceptable, even if the model has reached a T/CCL. Human editorial review is pending."},{"id":"evidence-ap-nate-soares-corrigibility-2015","sourceRole":"primary","support":"supports","locator":"PDF text layer, first two pages: author name on the title page or first page","passage":"Nate Soares","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-27","note":"Credited name checked by an automated research assistant on 2026-10-27 in a direct retrieval and checked by a reviewer from a different model family on 2026-10-27. Human editorial review is pending.","retrievalRoute":"direct"},{"id":"evidence-ap-eliezer-yudkowsky-sl4-2002-03-10-yudkowsky-0203-3160","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":"ev-cmp-faif-2026-d7-external-1","sourceRole":"primary","support":"supports","locator":"PDF p. 9, §3 Incident reporting, numbered item 4 (continued from PDF p. 8)","passage":"The edition says that, where applicable laws and regulations make an incident reportable, xAI will send the relevant authorities a copy of its incident report within the deadlines that apply.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-15","note":"Explicit paraphrase, written on 2026-10-27 by an automated research assistant from the comparison’s quotation of this passage, which was checked on 2026-10-15 against the publisher copy (SHA-256 recorded in the comparison pack). The edition could not be retrieved again on 2026-10-27; the archive capture of 2026-07-16 has the same SHA-256 (research/archive-checks-2026-10-26/codex-report.md). Given as a paraphrase to keep the edition within the quotation limit. Human editorial review is pending."},{"id":"ev-cmp-meta-faf-v1-1-d6-competitor","sourceRole":"primary","support":"supports","locator":"Meta FAF · v1.1: Whole edition, PDF pp. 1–21, read in the text layer of the decoded PDF of the Internet Archive capture of 2025-02-13; cover and PDF pp. 12–13 and 19 inspected visually","passage":"Within the inspected scope, no passage meeting the comparison question “Provisions conditioned on other developers” was found.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-27","note":"Absence recorded by a reviewer from a different model family on 2026-10-27, which read the whole edition in the Internet Archive capture of 2025-02-13, which it retrieved and hashed (decoded PDF SHA-256 9fa301df…cc21; research/frontier-framework-comparison/codex-meta-v1-1-2026-10-27.md) before opening the comparison’s stored cells. One reading; no second reviewer. Human editorial review is pending."},{"id":"evidence-fl-cn-genai-art7-training-data-1","sourceRole":"primary","support":"supports","locator":"Article 7 (第七条), chapeau","passage":"应当依法开展预训练、优化训练等训练数据处理活动","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: “shall carry out training-data processing activities such as pre-training and optimisation training in accordance with law”. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-cpo-dc-stargate-port-washington-capacity-1","sourceRole":"primary","support":"supports","locator":"paragraph 2","passage":"The campus will feature four cutting-edge data centers providing close to a gigawatt of AI capacity. Construction will begin soon and is scheduled for completion in 2028.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, compute, power and orbit (research/compute-power-orbit). 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 (SHA-256 0aea6b3ef87fcfa11f074347e414fcc95f3f6900c910440b8829929f9cb0fd50; the bytes differ from the pack’s retrieval). Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-fw-rsp-v2-2-cbrn-3-threat-model-rationale-1","sourceRole":"primary","support":"supports","locator":"PDF p. 23 (printed p. 19), Changelog, entry “May 14, 2025 (RSP v2.2)”, paragraph “ASL-3 Security”, third sentence","passage":"the CBRN-3 threat models entail large numbers of users having access to unguarded models (which is more likely to occur through a universal jailbreak than via model theft)","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (28 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-30, item 7) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-fw-rsp-v2-2-cbrn-3-threat-model-rationale-2","sourceRole":"primary","support":"supports","locator":"PDF p. 23 (printed p. 19), Changelog, entry “May 14, 2025 (RSP v2.2)”, paragraph “ASL-3 Security”, third sentence, second clause","passage":"the relatively small number of employees who might be capable of model theft does not significantly affect the risk level","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (20 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. Added after the audit of 2026-10-28: the second of the two reasons the sentence gives, cited so that the reason in the statement is not read as the only one. A cross-family machine check (2026-10-30, item 7) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-scr-igsc-hsp-v31-refusal-reporting-1","sourceRole":"primary","support":"supports","locator":"7.4 Order Refusal and Reporting, requirement 4.1, p. 15","passage":"IGSC members reserve the right to refuse to fill any order and to notify other IGSC members and/or law enforcement or other authorities upon identifying potentially problematic orders.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (28 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. A cross-family machine check (2026-10-30, item 75) re-read the record’s sources and found partial support; its corrections were applied at admission. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-scr-igsc-hsp-v31-refusal-reporting-2","sourceRole":"primary","support":"supports","locator":"7.4 Order Refusal and Reporting, requirement 4.3, p. 15","passage":"the reporting requirement covers requests within a category of sequences the protocol defines, received from a suspicious potential customer or from a potential customer failing to establish legitimacy under the protocol’s customer-screening practices.","passageType":"paraphrase","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The paraphrase is Selekoa’s, written from the text retrieved for the candidate pack on 2026-10-28; it is not a quotation and was not machine-matched. The paraphrase was narrowed at admission to the bounded policy scope, without the protocol’s own definition of the category (cross-family check, item 75). A cross-family machine check (2026-10-30, item 75) re-read the record’s sources and found partial support; its corrections were applied at admission. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-arg-d7-brent-articulates-1","sourceRole":"primary","support":"supports","locator":"Summary, p. iv (PDF p. 4)","passage":"Second, safety assessments rely on imperfect benchmarks that overlook the full range of potential threat actors, failing to account for how AI could enhance the capabilities of skilled individuals in addition to its ability to enable or \"uplift\" nonexperts.","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (39 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 paper is the primary source for its own argument. A cross-family machine check (2026-10-30, item 118) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"},{"id":"evidence-ab-arg-d7-brent-articulates-2","sourceRole":"primary","support":"supports","locator":"Submission history block, v1 entry","passage":"[v1] Thu, 12 Jun 2025","passageType":"quote","verification":"machine_checked","checkedAt":"2026-10-30","note":"Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (5 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval (SHA-256 63834668f90e3a0e80b98abc1ee48e42834a3ecc9060acc685a1a048bafb24e2; the bytes differ from the pack’s retrieval). Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. Added after the audit of 2026-10-28: the day of the date rested on no cited passage. Cited for the submission date only; the retrieved PDF carries no arXiv stamp and its cover gives only the month. A cross-family machine check (2026-10-30, item 118) re-read the record’s sources and found full support. No human editorial review.","retrievalRoute":"direct"}],"at":{"entities":[1,7,19,29,36,39,43,44,48,52,53,57,157,187,190,199,314,322,378,467,503,641,678,923,1177,1198,1206,1235,1260],"claims":[15,380,445,448,1185,1618,1638,1909,2088,2119,2327,2408,2690,2855,2902,3158,3271,3321,3434],"sources":[6,38,42,47,48,52,152,171,178,318,319,327,434,569,638,836,984,1050,1076,1077],"evidence":[15,403,477,480,1437,1438,2038,2039,2066,2471,2701,2702,2748,2974,3055,3378,3658,3742,4205,4485,4486,4561,4562,4737,4738]},"version":"1lg47t7"}