Anthropic Doubles Its AI Development Speed While CEO Calls to Slow Down

By Daniel Okafor

Anthropic is improving its AI models roughly twice as fast as it was a year ago, according to an analysis released Thursday by the independent monitoring group Model Evaluation and Threat Research. The finding landed weeks before the San Francisco company's Wall Street debut, which could rank as the biggest initial public offering of stock ever.

The timing is awkward. During METR's assessment period, Anthropic CEO Dario Amodei publicly urged the industry to rein in its pace. In a Sept. 12 essay responding to concerns about runaway AI, he called on the AI world to "slow the pace at which we improve the capabilities of AI models."

Anthropic did not immediately comment on the report in response to inquiries.

Why independent eyes on an AI lab matter

Anthropic invited the Berkeley, California-based nonprofit to review its internal data and interview staff as debate swirls over whether AI models could elude human control with catastrophic consequences. The findings carry weight because details about the technology and its risks have come mainly from AI companies themselves. President Donald Trump's administration has encouraged those firms to police themselves on safety.

METR has evaluated model capabilities for Anthropic, OpenAI and other labs for years. The nonprofit runs on donations and takes no payment for its work, though labs it partners with have provided access and tokens for evaluations, research and engineering.

One of its two reports drew on Anthropic's internal data over roughly 11 weeks, from mid-August through late October, plus employee interviews. METR said it focused on two questions: whether the growth rate of Anthropic's internal capabilities has accelerated, and how much of that was attributable to AI. The group said it received "unprecedented access to internal data" on Anthropic's research and development process.

Anthropic's redactions limited how METR could communicate more detailed estimates of acceleration, and METR's own team carried internal disagreements and uncertainty. Still, the group concluded the rate of improvement had doubled.

"Anthropic appears to have made as much progress on benchmarks in the past year as they did between October 2023 and October 2025," the METR report stated.

METR also compared Anthropic's public risk report from August against a private version and other non-public data. The organization said the public report "overall understates the level of acceleration that was occurring."

What is driving the acceleration?

METR estimated the speedup applies to "well specified tasks" and believes much of the cause is AI helping to build AI. Anthropic has said its Claude chatbot increasingly assists in developing newer, more advanced versions of itself.

The report carried caveats. METR said it could be overstating its findings, because the acceleration might be limited to tasks that can be graded automatically or with a language model, rather than "messy" tasks that resist definition. Alternatively, Anthropic might have understated its earlier progress, making the latest number look larger by comparison.

The researchers said they could "confidently" rule out growth in training compute as the explanation. Compute has traditionally tracked model performance closely. Their best guess: the improvement comes from using AI to improve AI.

A key concern, METR said, is that AI-driven self-improvement could compound the acceleration. That risk looks sharper because researchers found no "bottlenecks" — no restrictions on computing power or data — holding back Anthropic's progress.

"We understand that AI is not yet able to make highly significant and fully autonomous contributions to the most important parts of the model training process," the report said. But Anthropic researchers described significant "signs of life" suggesting that could happen soon, according to METR.

What did Amodei say about self-improving AI?

Amodei addressed the mechanism directly in his September essay. He said AI advances have been driven primarily by recursive self-improvement, which he defined as "AI's growing ability to build the next generation of AI."

"Left unchecked, it could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all," Amodei wrote.

In the same essay, he championed embedding external evaluators inside leading AI labs and named METR as an appropriate third-party checker. OpenAI CEO Sam Altman said at the time that he also supported independent evaluators with employee-like access.

The report now gives investors heading into Anthropic's IPO a rare outside measurement of the engine behind the company's valuation — one that shows the engine spinning twice as fast as its own CEO says is safe.