Anthropic data shows AI now leads 26% of its R&D, up from under 1% in six months

Anthropic has released new data showing its AI model Claude now "leads" 26% of the company's AI research and development, up from less than 1% in February, intensifying concerns about recursive self-improvement and long-term AI risks. The company defines "leading" as the AI completing most of a task end-to-end from a high-level prompt under human supervision. Over 90% of AI R&D is now performed at a level where AI "collaborates" with or surpasses humans. In August, roughly 30,000 AI agents ran simultaneously on Anthropic's main internal platform, with about 0.002% of over 1 billion agent decisions blocked by monitoring systems. Anthropic has not yet reached fully autonomous AI R&D, as humans remain in the process, but the rapid growth from under 1% to 26% in six months is notable. About 12% of AI-related computing power in July was directed toward safety work, though the company cautioned that computing power is an imperfect measure of safety investment.