Anthropic spoke on Thursday.

It said its AI model Claude helps build a smarter version of itself.

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

So the model can finish most of a task from a big prompt.

But humans still watch it.

Claude does not work fully alone yet.

About 90% of Anthropic's research and development happens with Claude.

The company said this.

It said the model does big parts of the work.

Humans direct it closely.

Claude's lead share grew fast.

It was zero in February.

It reached a quarter of the work in August.

Some AI leaders want to slow down the technology.

They worry about safety.

Anthropic CEO Dario Amodei is one of them.

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

It said this needs better measurement of AI development.

It also needs public reporting.

Then society can decide how to use that information.

Anthropic warned about fast self-improvement.

Models that speed up their own development can be hard for humans to understand.

They can be hard to control too.

Anthropic said sharing these numbers could show how close top labs are.

The goal is recursive self-improvement.

Then a model can build its successor alone.

Anthropic asked other developers to share similar numbers often.

They should use a public method.

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

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

It recently promised to add outside third-party evaluators to watch safety.