Papers & Research

METR: Anthropic’s capability progress roughly doubled, with key evidence heavily redacted

On Nov 5, 2026, independent evaluator METR published a high-access assessment at Anthropic (Aug 12–Oct 29) on whether AI is accelerating AI R&D itself. As of October 2026, Anthropic appears to improve on well-specified tasks at about twice the March 2024–October 2025 pace (METR’s uncertainty spans ~1.3×–3×). On Anthropic’s AECI index, growth since October 2025 is about 1.9× [1.5–2.8], roughly two years of historical progress in one year. METR is confident the speedup is not from faster training-compute scaling; most of today’s acceleration likely comes from staff using AI to assist R&D, not yet fully autonomous core training work, though “signs of life” for autonomy exist. It saw no clear hard bottlenecks in algorithms, train/experiment compute, or data over the next 12 months, while admitting that view changed several times. Anthropic redacted much of the most informative evidence in the ~100-page report, so the public must lean on METR’s judgment. This is distinct from METR’s same-day review of Anthropic’s August Risk Report R&D section already on our site.

Quick Facts

Published
Updated
Source
METR
Category
Papers & Research
Topic
Anthropic / ClaudeSafety & AlignmentIndustry News
Why it matters

少有的进厂独立评估:能力进步是否因 AI 自我加速,且与站内已有 METR 风险报告评审不是同一篇。

Topics

Related entities

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: METR: Anthropic’s capability progress roughly doubled, with key evidence heavily redacted. November 5, 2026. https://new.nat6.net/en/news/custom-727783c45e414d2e/

View all →