Anthropic's AI is improving twice as fast as a year ago, watchdog says

Anthropic is improving its AI systems roughly twice as fast as it did a year ago. That's the finding of an independent analysis released on Thursday, and it lands awkwardly: the company's own CEO has been asking the industry to slow down.

What happened?

Anthropic invited Model Evaluation and Threat Research (METR), a Berkeley, California nonprofit, to look at its internal data and interview its staff. The review covered about 11 weeks, from mid-August to late October.

METR estimates that Anthropic is improving its models' ability to do "well specified tasks" around twice as fast as a year earlier. Its report says the company made as much progress on benchmarks in the past year as it did between October 2023 and October 2025.

The group says it can "confidently" rule out more computing power as the reason. Its best guess is that AI itself is helping build better AI. Anthropic has said its Claude chatbot increasingly helps develop newer versions of itself.

Why is everyone talking about it?

Most of what we know about AI risk comes from the companies building it. They have been encouraged by the Trump administration to police themselves on safety, so outside access like this is rare.

METR says it got "unprecedented access". It also says Anthropic's public risk report from August "overall understates the level of acceleration" that was happening.

The researchers found no "bottlenecks", such as limits on computing power or data, holding progress back. Their worry is that AI improving AI could make the speed-up compound. They add that AI can't yet make fully autonomous contributions to the most important parts of training, though Anthropic's researchers described "signs of life" that it could soon.

There are caveats. The speed-up may only apply to tasks that are easy to grade, not "messy" ones. And Anthropic's earlier progress may have been understated, which would make the new figure look bigger. Anthropic's redactions also limited what METR could say.

What's next?

Co-founder Sam McCandlish said on X that many of METR's conclusions match his own. "AI is speeding up our research, but it's hard to say exactly how much," he wrote. Anthropic plans to publish its own analysis in the coming weeks.

He defended the redactions, saying details of what drives the speed-up are "commercially sensitive" and could hand others "a roadmap". The company calls that a tradeoff.

For context, CEO Dario Amodei wrote in a Sept. 12 essay that the industry should "slow the pace at which we improve the capabilities of AI models". Anthropic is also planning to go public later this month, in what could be the biggest IPO ever.