Anthropic is using AI to build more powerful AI, alarming researchers

A report published Thursday by METR, the AI safety research organization, found that Anthropic’s internal pace of AI research and development has roughly doubled in speed compared to a year ago, with little sign of slowing. The findings follow an eleven-week investigation conducted inside the company.

According to the report, Anthropic is increasingly using its own AI systems to accelerate the development of subsequent, more capable models. METR’s evaluation examined how this automation is reshaping the company’s research and development pipeline, tracking the rate at which new capabilities are produced and deployed.

The Verge, summarizing the report’s release, notes that the trend has drawn concern from researchers tracking AI safety, given the implications of a compounding feedback loop in which AI systems contribute directly to designing their successors. The report does not claim the trend will continue indefinitely, but it documents sustained acceleration over the period studied, with no observed deceleration.

Anthropic CEO Dario Amodei was referenced in coverage of the report, though the summarized material does not include a direct statement from him addressing METR’s specific findings.

The report represents an external, independent look inside a leading AI laboratory’s development practices, offering one of the more detailed outside accounts to date of how automation is altering the speed of frontier AI research.