Independent Researchers Find Anthropic's AI Development Is Accelerating at Twice Its Previous Pace
An independent assessment by research organization METR found that Anthropic is advancing AI capabilities at roughly twice the rate seen in an earlier period, with AI tools helping employees work faster on coding and research tasks. Researchers warn that much of the supporting evidence was redacted by Anthropic.
An independent assessment published November 5, 2026, found that Anthropic is advancing its AI capabilities at approximately twice the pace seen during an earlier measurement period, with AI tools playing a central role in the acceleration.
The research organization METR examined internal development data from Anthropic spanning August 12 through October 29. Researchers found evidence that AI tools are helping Anthropic employees work faster on tasks including coding and analyzing experiments.
The finding raises questions about recursive self-improvement, a process in which AI helps researchers build better models, which could then develop their successors even faster. Most of the acceleration METR observed still came from AI assisting people rather than independently conducting entire research projects. But researchers found no clear bottleneck that would necessarily prevent further acceleration over the next year.
The public cannot examine much of the evidence behind the conclusions. METR said Anthropic redacted most of the material it considered most informative in its roughly 100-page assessment. Anthropic said the redactions were intended to protect confidential information that could help competing AI developers.
METR could publish its overall conclusions, but many underlying measurements and forecasts remained confidential. The researchers are warning that AI progress may be speeding up while asking the public to trust conclusions it cannot fully verify.
The assessment comes as the broader AI industry faces growing scrutiny over data center expansion, environmental impact, and the pace of capability development. Political resistance to AI infrastructure intensified after the U.S. midterm elections, with governments from New York to Amsterdam tightening restrictions on data center development.
The next question, METR said, is whether increasingly capable models will begin completing larger portions of AI research independently, and whether outside evaluators will have enough access to assess the consequences.


