Anthropic’s debate over the pace of AI development is becoming harder to separate from its own record.

Between March 2024 and October 2025, the company established the baseline against which the new assessment measured its progress. In mid-August, the AI-safety organization METR began communicating with Anthropic as it examined whether the company’s research process was itself being accelerated by artificial intelligence.¹

By mid-September, CEO Dario Amodei had publicly staked out the opposite instinct: frontier AI should be advanced deliberately. His essay, “We Must Pace the Frontier,” framed restraint as a necessity rather than a retreat.¹

METR’s final public report, submitted on November 3, points to a striking countercurrent. The group said Anthropic was “likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025.”¹ It judged that the faster pace was “likely” tied to Anthropic’s extensive use of AI in its own research and development.¹

A separate account described the result in similarly stark terms: Anthropic was developing AI systems “around twice as quickly as it was a year ago,” even as Amodei called for a slowdown in the technology’s rate of improvement.²

The two accounts do not allege that Anthropic abandoned its safety case. Instead, they expose the central contradiction facing frontier labs: the tools meant to be governed cautiously may also be making their creators markedly faster at building the next generation.