Anthropic said on Thursday that its AI model Claude is helping to build a smarter version of itself.

The company said Claude now leads 26% of its model research and development.

This means the model can finish most of a task from a simple, high-level prompt.

But humans still watch over it.

Claude does not yet work fully on its own.

About 90% of Anthropic's research and development happens together with Claude, the company said.

It described this as the model doing large parts of the work under close human direction.

The share of work Claude leads has grown fast.

It was zero in February, and it reached a quarter of research and development work in August.

The news came as some AI leaders call for slowing down the technology over safety concerns.

This group includes Anthropic CEO Dario Amodei.

Anthropic said frontier labs and the public should not have a large gap in what they know.

It said this requires better measurement of AI development and public reporting.

Then society can decide how to use that information.

Anthropic warned that models speeding up their own development could make it harder for humans to understand or control them.

It said sharing these metrics could show how close leading labs are to recursive self-improvement.

That is when a model can build its successor on its own.

Anthropic urged other developers to share similar metrics regularly using a public method.

It did not say how close it believes it is to that goal.

Anthropic also said it had about 30,000 agents doing research and engineering work as of August.

It recently committed to adding outside third-party evaluators to monitor safety.