Anthropic's Claude AI Now Handles 26% of Its Own Development

Anthropic disclosed Thursday that Claude, its flagship AI model, now leads 26% of the company's model research and development work — a significant milestone in AI systems contributing to their own evolution.

The AI lab defined "leading" as Claude's ability to complete most of a given task end-to-end from a high-level prompt while remaining under human supervision. The model is not operating autonomously. Additionally, Claude participates in roughly 90% of Anthropic's research and development in what the company calls "collaboration," handling large portions of work under close human direction.

The disclosure comes amid growing debate over AI safety, with Anthropic CEO Dario Amodei among those calling for measured development of the technology.

Why it matters

Recursive self-improvement — when AI models can autonomously build their successors — represents a potential inflection point in artificial intelligence development. Anthropic's transparency about Claude's increasing contribution to its own advancement provides rare visibility into how quickly leading labs are approaching this threshold. The company's call for standardized public reporting could establish precedent for an industry that has historically operated behind closed doors on capability benchmarks.

Rapid acceleration in six months

The pace of Claude's expanding role is striking. In February, the model led none of Anthropic's research and development work. By August — just six months later — it had reached the 26% benchmark, according to details first reported by ABC7 Chicago.

Anthropod also revealed it operates approximately 30,000 AI agents conducting research and engineering work as of August. The company emphasized that oversight measures are essential for detecting agent misbehavior through monitoring systems.

Push for industry standards

Anthropod argued in a blog post that "we should do everything possible to minimize the gap between what frontier labs know and what the public knows." The company urged competing AI developers to share similar metrics regularly using public methodologies that would allow comparisons over time and across laboratories.

The company acknowledged that models accelerating their own development could make it "more challenging for humans to understand or control these systems." Sharing standardized metrics could help society understand how close leading labs are to achieving recursive self-improvement and make informed decisions about the technology's trajectory.

Safety measures and external oversight

Anthropod recently committed to embedding external third-party evaluators within the company to monitor safety efforts. This follows broader industry turbulence around AI safety concerns.

An Anthropic researcher recently resigned with warnings about AI threats to humanity, sparking renewed dialogue. While Amodei, OpenAI CEO Sam Altman, and Elon Musk have supported slowing development, other technology leaders and President Donald Trump have pushed back against such proposals.

Anthropod's disclosure did not clarify how close the company believes it is to achieving full recursive self-improvement, but the trajectory suggests rapid progress toward AI systems that can meaningfully contribute to building their own successors.

These details were first reported by ABC7 Chicago.