report · 2026-11-05
Review of the "Risks from automated R&D" section in Anthropic's Risk Report (August 2026)
On this page
Why it matters here
METR's November 2026 review distinguishes Anthropic's AI R&D early-warning threshold from full substitution of researchers and from direct catastrophic risk. For models covered through July 15, METR broadly agrees with the low direct-risk and non-substitution conclusions but considers crossing the acceleration warning threshold plausible and the public report's confidence overstated. It identifies omitted uncertainty and contestable choices about leading indicators, AI attribution and a one-time capability jump versus sustained acceleration.
AI-R&D capabilityEvaluation integrity
What to keep in mind
- This is a historical, attributed external review of the August 2026 report and July 15 model coverage, not a current tracker risk verdict or proof that a threshold was crossed. Most relevant METR and Anthropic researchers reportedly lean toward the warning threshold not having been crossed.
- METR had private evidence that this tracker cannot inspect. Public Section 3.5 was cross-checked, but the private quotations, leading indicators, counterfactual calculations and full assessment cannot be independently verified.
- Early warning, researcher substitution, individual productivity, benchmark growth and acute catastrophic risk are distinct claims. No reported multiplier is converted into a comparable numerical observation or a demonstrated recursive improvement loop.
- The review shares the assessment's mediated access, redaction rights, free resources, use of Claude, conflicts and informal publication/access pressures. The full public report and disclosures remain the source of these qualifications.
- Checks here cover the review's main text/footnotes, shared process disclosures and the original public risk report's relevant threshold/Section 3.5 text. The 186-page report's other evaluations, figure pixels and cited experiments were not audited or adopted.