Why a new report on Anthropic automating AI development is alarming researchers

Why it matters

A METR investigation inside Anthropic found that the company's AI models are improving markedly faster than they were a year ago, and that most of this acceleration is attributable to AI itself rather than expanded computing resources. The safety organization conducted an extended probe with access to confidential research, performance data, and employee interviews, and reported that staff using Anthropic's most capable internal model described themselves as far more productive than they would be without AI assistance. METR also disclosed that Anthropic has finished training several unreleased models in recent months, suggesting rapid internal development beyond what the public sees. The report's significance is tempered by Anthropic's redaction rights: the company removed a majority of the evidence METR considered most informative, citing competitive concerns, which limits public ability to judge the findings. The situation is notable for the tension it reveals, as Anthropic prepares to go public while its CEO, Dario Amodei, has urged labs to slow AI development even as AI accelerates the company's own research.