Anthropic AI development speed nearly doubles, METR audit finds

An audit by the non-profit organization METR has revealed that the development speed of Anthropic’s AI models has approximately doubled. According to the findings, about 80 percent of this acceleration is attributed to the use of AI within the company’s own research processes.

Data from the 11-week evaluation, conducted between August and October 2026, shows that performance on specific tasks has increased 1.9 times faster since October 2025 compared to the period between March 2024 and October 2025. This acceleration is measured using a composite metric called AECI.

Experts warn that this trend could lead to a recursive self-improvement loop, where AI models develop subsequent models at increasing speeds. Such a cycle is cited as a potential risk for the sudden emergence of extreme and uncontrollable capabilities. While some internal surveys suggested productivity increases of up to 12 times due to AI assistance in coding and experimentation, METR noted that these specific figures lack definitive proof.

This audit follows a commitment by Anthropic CEO Dario Amodei to allow external auditors access to the company. Meanwhile, Anthropic is reportedly preparing for an initial public offering expected in November.