An outside research group with access to Anthropic’s internal development process estimates that the company’s artificial-intelligence capabilities have been improving almost twice as quickly as they were a year earlier.

Model Evaluation and Threat Research, or METR, reviewed roughly 11 weeks of internal data, interviewed staff and examined information about unreleased systems. Reuters reported that the nonprofit estimated Anthropic’s benchmark capabilities had improved 1.9 times faster since October 2025 than during the earlier period it studied.

METR estimated that the use of AI in research and development may account for about 80% of the acceleration. The group emphasized that the figure is uncertain and could overstate what is happening across the company’s broader work.

The review focused heavily on tasks that are clearly defined and can be graded automatically or by another model. Results from those tasks may not carry over to less structured work that requires scientific judgment, long-range planning or hard-to-measure creativity.

The researchers said they found no convincing evidence that slower growth in computing power or other major bottlenecks would necessarily halt the pace of improvement over the next year. At the same time, they found that AI systems are not yet making fully autonomous, highly significant contributions to the most important stages of model training.

That distinction limits what can be concluded from the report. The findings suggest AI tools are helping researchers build newer systems more quickly, but they do not establish that Anthropic’s models can independently create and train their successors or that rapid recursive self-improvement has begun.

Anthropic announced in October that six METR employees had received company hardware, office space and access to internal tools roughly comparable to the access given to its own risk-assessment teams. Customer data, attorney-client privileged material and certain architecture information remained outside the arrangement.

Anthropic said it could require defined categories of confidential information to be redacted but could not veto publication of METR’s findings. METR may publicly describe its access and challenge whether requested redactions were justified without disclosing the protected material.

The review arrives as policymakers and technology companies debate whether independent evaluators can provide credible oversight of frontier AI systems. It also comes as Anthropic prepares for a proposed public stock offering, increasing scrutiny of both the company’s commercial prospects and the risks associated with faster development.

Sources: Associated Press — Anthropic development acceleration and METR review: https://apnews.com/article/9cf8c590ffc1abdb489179bdccc69c23 Reuters — Anthropic models improving nearly twice as fast: https://www.reuters.com/technology/anthropics-models-improving-nearly-twice-fast-with-help-ai-study-finds-2026-11-05/ Anthropic — Update on independent evaluators: https://www.anthropic.com/news/an-update-on-independent-evaluators