Anthropic’s Claude models now lead 26% of the company’s artificial intelligence research and development work, a sharp rise from less than 1% in February 2026. The company disclosed the figure as part of a new effort to track how AI is being used to develop future AI systems.
The measurement covers data from August 2026. Anthropic said more than 90% of its measured AI R&D work is now performed at a level where AI either collaborates with humans or leads the task. Claude is still not fully autonomous in any measured area of the company’s R&D work.
Anthropic uses an automation scale developed by Epoch AI to assess the role of AI in research tasks. The ‘AI leads’ level means the system can complete most of a task from a high-level instruction while a human supervises the work.
The company’s chart shows that Claude was at the “leads” level for less than 1% of its AI R&D work in February. That share reached 26% by August. Anthropic said the figure does not mean Claude independently develops 26% of its next AI model. Human supervision is still part of the process.
Anthropic said, “Claude is not operating fully autonomously for any measured subset of AI R&D work.”
The company also reported that around 30,000 AI agents were carrying out research and engineering work at any one time on its most-used internal platform in August. Their actions were monitored through online and offline systems.
Anthropic analysed more than one billion agent decisions during August. Its online monitoring system blocked 0.002% of those decisions, equal to about one in 47,000. Around 100,000 transcripts were flagged each week for further review, with about 50 high-priority cases sent to human reviewers. Anthropic also said about 6% of its AI R&D computing resources went towards safety work during July 13-20. The share was about 12% for computing used in AI-driven R&D.
The company said these measurements are intended to give researchers, governments and the public a clearer view of how quickly AI development is becoming automated. Anthropic also called for other frontier AI developers to publish comparable data using methods that can be independently verified.
Also Read: Claude Gets One-Window Upgrade as Anthropic Brings Cowork Into Chat