Claim and evidence Close ×
Claude 3.7 Sonnet System Card states evaluation result ↓ The card states that in one uplift study models provided some amount of uplift to novices compared with participants without model access, and that the highest scoring plan from a participant with model access still included critical mistakes that would lead to failure in the real world. 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. Copy claim link
Copy claim reference
Machine-checked, not reviewed by an editor Stated in source Supported
Supporting evidence: primary · 1 source; Context: primary · 1 source
Date Date unknown Evaluated system Claude 3.7 Sonnet and, where the card says so, a helpful-only version of it What the report measured Whether access to a model increased what participants could produce on tasks proxying real-world CBRN tasks By what kind of method A human-participant uplift study with model-access and no-model-access conditions, scored against the study's own criteria The result, in the report’s words 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 The caveats the report gives 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. Scope and limits One study as the developer reports it, for the models and participants it describes. The record does not say how large any uplift was, does not generalise it to other systems or settings, and carries the card's own statement that plans still failed on critical points. The study's tasks and the card's description of where performance improved are technical content and are excluded; only the stated comparative outcome is recorded.
The evidence 03 primary source Supports the claim
Claude 3.7 Sonnet System Card ↗ Anthropic · Publication date unknown
Copy source reference
Cite the original source Cite the original document, not Selekoa. Check these details against it.
Copy APA
Download BibTeX Download RIS Download CSL-JSON
Source excerpt “in one uplift study, we found that models provided some amount of uplift to novices compared to other participants who did not have access to a model” Location in source PDF p. 23, §7.1 CBRN Evaluations, 'Results' paragraph
How it was read Retrieved directly from the publisher
Machine-checked · 2026-10-30
Source details and check notes Authors Anthropic 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). Canonical address Open the canonical source ↗ Accessed 2026-10-28 Language English (en) Attribution Anthropic, “Claude 3.7 Sonnet System Card”. Reuse No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution. Check notes 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. A cross-family machine check (2026-10-30, item 50) re-read the record’s sources and found full support. No human editorial review. primary source Context only
Claude 3.7 Sonnet System Card ↗ Anthropic · Publication date unknown
Copy source reference
Cite the original source Cite the original document, not Selekoa. Check these details against it.
Copy APA
Download BibTeX Download RIS Download CSL-JSON
Source excerpt “even the highest scoring plan from the participant who had access to a model still included critical mistakes that would lead to failure in the "real world".” Location in source PDF p. 23, §7.1, 'Results' paragraph, following sentence
How it was read Retrieved directly from the publisher
Machine-checked · 2026-10-30
Source details and check notes Authors Anthropic 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). Canonical address Open the canonical source ↗ Accessed 2026-10-28 Language English (en) Attribution Anthropic, “Claude 3.7 Sonnet System Card”. Reuse No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution. Check notes 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. A cross-family machine check (2026-10-30, item 50) re-read the record’s sources and found full support. No human editorial review. primary source Context only
Claude 3.7 Sonnet System Card ↗ Anthropic · Publication date unknown
Copy source reference
Cite the original source Cite the original document, not Selekoa. Check these details against it.
Copy APA
Download BibTeX Download RIS Download CSL-JSON
Source excerpt “how the model performs on expert-level evaluations may not directly translate to real-world capabilities” Location in source PDF p. 23, §7.1, 'Results' paragraph
How it was read Retrieved directly from the publisher
Machine-checked · 2026-10-30
Source details and check notes Authors Anthropic 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). Canonical address Open the canonical source ↗ Accessed 2026-10-28 Language English (en) Attribution Anthropic, “Claude 3.7 Sonnet System Card”. Reuse No open redistribution licence asserted. Bibliographic metadata, editorial paraphrases and short attributed quotations (at most 50 words each). Recheck terms before wider redistribution. Check notes Candidate pack passage, AI and biosecurity (research/ai-bio). The quotation (14 words) matched the text retrieved for the candidate pack on 2026-10-28 and again at admission on 2026-10-30, in a fresh retrieval whose SHA-256 equals the pack’s. Matching ignored differences of whitespace, case, quotation marks, dashes and Unicode compatibility forms, so the source’s typography can differ. The card's own limit on reading its evaluation results. A cross-family machine check (2026-10-30, item 50) re-read the record’s sources and found full support. No human editorial review. Status and meaning Editorial status. Not reviewed by an editor . An exploratory record that no editor has accepted. This status does not say whether a person has read any linked passage. Each evidence record above shows how it was checked.
Relation. A report states the result of an evaluation it ran or commissioned, in its own words, with what was measured, by what kind of method, and the caveats the report itself gives (schema 2.22). The record is the report’s statement, not a verified finding: it says nothing about another system, a later version or real-world risk, and a null or limiting result is not evidence that no effect exists.
Date. The card gives no date for the study or for its own publication.
Record ID ab-ev-claude37-uplift-study-result