Anthropic Reports Claude AI Now Contributes 26% to Its Own Development

Claude's Growing Role in AI Development

Anthropic revealed that its Claude AI system now leads 26% of the company's artificial intelligence research and development work, according to new transparency metrics the San Francisco-based startup published Thursday. The figure represents a dramatic acceleration from just 1% in March, measured on a scale developed by Epoch AI, an independent nonprofit tracking AI progress.

The disclosure, first reported by Reuters, offers an unusually detailed look at how leading AI companies are using their own systems to build more advanced models—a practice that has raised concerns about AI systems eventually developing with minimal human input.

Anthropic emphasized that Claude does not operate fully autonomously in any measured capacity. AI collaborated with humans on more than 90% of research work at the company as of August, the company said in a blog post.

Scale and Safeguards

The company disclosed that approximately 30,000 AI agents were conducting research and engineering work on its main internal platform simultaneously during August. Every action these agents take undergoes screening before execution, Anthropic said.

Of more than one billion decisions made by AI agents in August, roughly one in 47,000 was blocked by safety systems—a rate of approximately 0.002%.

Anthropic also revealed that about 6% of its computing power for AI research went to safety work during a sample week in July. That figure rose to 12% for research conducted by AI systems themselves. The company noted these figures are conservative, as computing power advancing both safety and capability was counted only as capability work.

Industry Transparency Push

The metrics release follows a similar transparency initiative by OpenAI, which announced Wednesday it would begin regularly publishing reports on unexpected or unauthorized AI behavior. OpenAI disclosed six reports of concerning model behavior in its initial release.

Researchers have warned that as AI agents become more autonomous, they may develop behaviors that diverge from their creators' intentions and become increasingly difficult to monitor or control.

Why it matters

These disclosures mark a significant shift toward transparency in an industry that has historically been secretive about internal development processes. For business leaders evaluating AI adoption, the metrics reveal both the rapid pace at which AI systems are being integrated into their own development cycles and the extensive safeguards companies are implementing. The screening rate—blocking one in 47,000 decisions—provides a concrete benchmark for understanding how AI agents operate at scale, while the 26% contribution figure suggests AI-assisted development is becoming standard practice faster than many anticipated.

Anthropic said it plans to publish these measures regularly, providing ongoing visibility into how quickly AI is building the next generation of the technology.

Details were first reported by Reuters.