METR, the AI evaluation nonprofit, published on November 5, 2026 an independent assessment of how much AI is speeding up AI research and development inside Anthropic. Authors Tom Cunningham, Nate Rush, Parker Whitfill and Hjalmar Wijk report on work from August 12 to October 29, 2026, when METR staff held high access to Anthropic internal metrics, interviews and documents. The engagement is separate from Anthropic's October 15 program that gives METR employee-like access for frontier risk work.
METR's headline judgment is that as of October 2026 Anthropic is likely improving model capabilities on well-specified tasks at roughly twice the rate seen between March 2024 and October 2025, with wide uncertainty that could put the true figure nearer 1.3x or 3x. On Anthropic's AECI benchmark index, scores have grown at an estimated 1.9x the pre-October 2025 rate (95% confidence interval 1.5 to 2.8x), which METR reads as about two years of historical progress in one year. METR says it is confident the speedup is not from faster training-compute scaling.
Attribution is more uncertain. METR guesses that a majority of the current acceleration, perhaps around 80%, comes from Anthropic's extensive use of AI for R&D, while late-2025 acceleration likely mixed in non-AI factors that Anthropic redacted. Staff Slack polls and interviews point to large self-reported productivity gains, reaching about 12x with the most capable internal model in October 2026, though METR treats those polls as loose evidence. Researchers told METR that help still mainly assists humans rather than fully autonomous contributions, with signs that more autonomous work may arrive soon.
Anthropic redacted a majority of the evidence METR found most informative for its risk conclusions, citing commercial sensitivity and a wish not to hand rivals a roadmap. METR says the public therefore has to lean more on METR's judgment than would be ideal, and that much redacted material admitted multiple interpretations. The public write-up also notes three unreleased models, codenamed Model 2, Model 3 and Model 4, that finished training between July and October 2026, plus an unreleased 2022 model. Anthropic cancelled a planned compute-allocation verification midway, saying it wanted to keep the engagement focused on risk assessment for now.
Independent coverage arrived quickly. A short Verge note by Jay Peters on November 5 flagged the report; Robert Hart's longer Verge analysis on November 7 framed researcher alarm about recursive self-improvement timelines and Anthropic's redactions; AP also reported METR's roughly 2x finding. Anthropic cofounder Sam McCandlish wrote that heavy redactions were deliberate and that Anthropic plans its own progress-speed report. METR updated its AEF-1 checklist notes on November 6.
For AI desks the assessment matters because a high-access independent evaluator now says a frontier lab's capability progress has roughly doubled and that AI itself is likely the main driver, while bottlenecks that would stop further compounding were not clearly seen. It sits beside Anthropic's amended S-1 warning that it may soon meet its own AI-accelerated R&D threshold, but it is a distinct event: an outside measurement of acceleration, not a securities disclosure.