Independent assessment: Anthropic's AI research pace may have doubled
From the AI digest for Nov 5, 2026
What happened: METR, the AI research nonprofit that evaluates frontier labs, published an assessment of Anthropic built on a high-access evaluation running August 12 to October 29, 2026. Its core finding: Anthropic is likely improving its models on well-specified tasks at roughly twice the rate it did between March 2024 and October 2025 — its internal benchmark scores have grown at an estimated 1.9 times the pre-October 2025 rate (95% confidence interval: 1.5–2.8x), roughly two years of historical progress in the last year — and a majority of that acceleration is likely attributable to Anthropic's extensive use of AI to do AI research rather than to faster compute scaling. METR found no significant bottlenecks in algorithmic progress, training compute, experiment compute or data over the next 12 months, and warns that without bottlenecks, further capability progress could accelerate faster still. Anthropic redacted a majority of the evidence METR found most informative — described as removing confidential material that could accelerate other AI developers — which METR says leaves the public relying heavily on its judgment. In a companion review of the "Risks from automated R&D" section of Anthropic's August 2026 public Risk Report, METR agrees the July 15 models posed low catastrophic risk, but says the public report "understates the level of acceleration that was occurring" and that it is plausible Anthropic's 2x-acceleration early-warning threshold had already been crossed — a claim Anthropic's report disputes on what METR calls contestable judgment calls.
What's genuinely new: the first quantified, high-access external estimate that AI is accelerating AI development inside a frontier lab — with the open caveat that much of the supporting evidence is redacted — plus a direct external challenge to Anthropic's public claim that its early-warning threshold had not been crossed.
Watch: Anthropic's response to the assessment and the threshold critique; whether other labs commission comparable high-access evaluations; whether METR's 2x rate estimate holds, accelerates or reverses as newer data lands.
Links: METR · METR · The Information · Nov 5, 3:00 AM ET