Anthropic says Claude now leads a quarter of work building its next AI models

Anthropic reported that its AI model Claude now leads 26% of the research and development work for its next-generation AI systems, according to new metrics the company will publish regularly. The San Francisco-based startup said Claude collaborated with humans on over 90% of research work as of August, up from 1% in March on a scale by Epoch AI. Claude is not fully autonomous in any measured work, and every agent action is screened, with about one in 47,000 decisions blocked. The disclosure offers rare public insight into AI self-development amid concerns about autonomy and control. Anthropic also said about 6% of its computing power went to safety work, rising to 12% for AI-led research, while rival OpenAI announced plans to publish reports on unexpected AI behavior.