Anthropic says its Claude artificial intelligence system is now leading 26% of the company’s AI research and development work, highlighting how rapidly AI models are becoming involved in the development of newer generations of technology.
The figure was disclosed Thursday as part of a new measurement framework that Anthropic plans to use to provide greater visibility into how AI is contributing to its own development. According to the company, the 26% figure applies to work measured as of August 2026.
Anthropic said Claude’s role has expanded sharply in recent months. The company reported that the model was leading only about 1% of its AI R&D work in March, compared with 26% in August. The assessment uses an automation scale developed by Epoch AI, an independent nonprofit that tracks progress in artificial intelligence.
Human Oversight Remains in Place
Anthropic emphasized that Claude is not operating fully autonomously in any of the areas covered by its measurement. The company defines “leading” as Claude being capable of completing most of a task from a high-level instruction, while a human continues to supervise the work.
The company also said AI collaborated with humans on more than 90% of its research and development activities as of August. Under this category, AI systems can complete substantial portions of a task but remain under closer human direction.
The growing role of Claude comes as AI companies increasingly use their own systems for activities such as coding, experimentation, evaluation and model development.
Focus on Recursive Self-Improvement
Anthropic said the measurements are intended to help the public understand how quickly AI systems are becoming involved in building more capable AI. The company is particularly interested in tracking progress toward what researchers call “recursive self-improvement,” a scenario in which an AI system could eventually help design and develop its own successor with minimal human involvement.
Anthropic stressed that such fully autonomous development has not been reached. It also acknowledged that measuring AI involvement remains challenging and that independent verification could improve comparisons across AI developers.
The disclosure arrives amid broader industry discussions about the pace of AI development and the safeguards needed as models take on increasingly sophisticated tasks. Anthropic said it intends to publish these measurements regularly, giving researchers and the public a way to track changes in AI-led development over time.
Also Read :- Anthropic’s Revenue Run Rate Hits $65 Billion, Boosting Amazon and SpaceX
