{"entities":[{"id":"anthropic","name":"Anthropic"},{"id":"claude-3-7-sonnet-system-card","name":"Claude 3.7 Sonnet System Card","subtype":"Technical report","description":"The 42-page PDF expression served from assets.anthropic.com, inspected 2026-10-28. The card states no publication date; its references are dated February 2025. No date is asserted.","sourceIds":["source-ab-claude37-card"]}],"claims":[{"id":"ab-ev-claude37-asl2-determination","statement":"The Claude 3.7 Sonnet System Card states that ASL-2 safeguards remain appropriate for the model, that its evaluations suggest improved performance in all domains including some uplift in CBRN evaluations, and that on what it observed in recent CBRN testing there is a substantial probability that the next model may require ASL-3 safeguards.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-claude37-asl2-determination-1","evidence-ab-ev-claude37-asl2-determination-2","evidence-ab-ev-claude37-asl2-determination-3","evidence-ab-ev-claude37-asl2-determination-4"],"attributes":{"evaluatedSystem":"Claude 3.7 Sonnet, final release model unless the card states otherwise","evaluation":{"what":"Whether the model reaches the ASL-3 capability threshold for CBRN, with biological risks the primary focus","how":"Automated knowledge and skill-testing evaluations, uplift studies designed to proxy real-world tasks, external expert red teaming and long-form agentic evaluations","result":"ASL-2 safeguards remain appropriate; improved performance in all domains, including some uplift in CBRN evaluations; a substantial probability that the next model may require ASL-3 safeguards","caveats":"The card states that some specific evaluation thresholds and results in the CBRN domain are considered sensitive and are not shared publicly, and that how the model performs on expert-level evaluations may not directly translate to real-world capabilities."}}},{"id":"ab-ev-claude37-card-published","statement":"Anthropic published the Claude 3.7 Sonnet System Card, which states that its Responsible Scaling Policy evaluations cover chemical, biological, radiological and nuclear risk among the areas of potential catastrophic risk.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-claude37-card-published-1"]},{"id":"ab-ev-claude37-chemical-not-evaluated","statement":"The Claude 3.7 Sonnet System Card states that Anthropic does not currently run specific evaluations on chemical risk directly because it prioritises biological risks, and that it informs its views through chemical risk analysis performed by the UK and US AI institutes.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-claude37-chemical-not-evaluated-1"],"attributes":{"evaluation":{"what":"Chemical risk, as distinct from biological risk, in the card's CBRN evaluations","how":"No direct evaluation; the card states it relies on partners and on analysis by the UK and US institutes","result":"No specific evaluations on chemical risk are run directly, because biological risks are prioritised","caveats":"The card states that some mitigations for chemical risks are implemented even though no direct evaluations are run."}}},{"id":"ab-ev-claude37-uplift-study-result","statement":"The Claude 3.7 Sonnet System Card states that in one uplift study models provided some amount of uplift to novices compared with participants who did not have access to a model, and that even the highest scoring plan from a participant with model access still included critical mistakes that would lead to failure in the real world.","derivation":"explicit","resolution":"supported","editorialState":"seed","evidenceIds":["evidence-ab-ev-claude37-uplift-study-result-1","evidence-ab-ev-claude37-uplift-study-result-2","evidence-ab-ev-claude37-uplift-study-result-3"],"attributes":{"evaluatedSystem":"Claude 3.7 Sonnet and, where the card says so, a helpful-only version of it","evaluation":{"what":"Whether access to a model increased what participants could produce on tasks proxying real-world CBRN tasks","how":"A human-participant uplift study with model-access and no-model-access conditions, scored against the study's own criteria","result":"Some amount of uplift to novices compared with participants without model access; the highest scoring plan with model access still included critical mistakes that would lead to failure in the real world","caveats":"The card states that results differed across its evaluation strategies, with some showing uplift and others not, and that expert-level evaluation performance may not directly translate to real-world capabilities."}}}],"sources":[{"id":"source-ab-claude37-card","title":"Claude 3.7 Sonnet System Card","authors":["Anthropic"],"publisher":"Anthropic","publicationDate":null,"version":"Retrieved for the candidate pack on 2026-10-28 (SHA-256 82947a46fc2d1b9a1d69501ef973fc3c5b328b703b9a17bd8a143c2672f8a1a6). 42-page PDF as served on 2026-10-28; the request to assets.anthropic.com redirected to https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf. The document itself states no publication date; its references are dated February 2025. Retrieved again at admission on 2026-10-30 (SHA-256 82947a46fc2d1b9a1d69501ef973fc3c5b328b703b9a17bd8a143c2672f8a1a6, the same bytes).","evidenceUrl":"https://assets.anthropic.com/m/785e231869ea8b3b/original/claude-3-7-sonnet-system-card.pdf","language":"en"}],"evidence":[{"id":"evidence-ab-ev-claude37-asl2-determination-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-asl2-determination-2","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-asl2-determination-3","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-asl2-determination-4","support":"context","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-card-published-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-chemical-not-evaluated-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-uplift-study-result-1","sourceRole":"primary","support":"supports","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-uplift-study-result-2","support":"context","verification":"machine_checked","retrievalRoute":"direct"},{"id":"evidence-ab-ev-claude37-uplift-study-result-3","support":"context","verification":"machine_checked","retrievalRoute":"direct"}],"at":{"entities":[1,1236],"claims":[3404,3405,3406,3408],"sources":[1081],"evidence":[4683,4684,4685,4686,4687,4688,4690,4691,4692]},"version":"1lg47t7"}