Anthropic said Claude led 26% of its measured artificial-intelligence research and development work in August, up from 1% in March under a framework developed by the independent nonprofit Epoch AI. Claude was not fully autonomous in any measured area; the figure reflects work in which AI took the leading role alongside human researchers. More than 90% of Anthropic’s AI R&D was conducted at the AI-collaborates level or higher. Anthropic said about 30,000 AI agents were conducting research and engineering work on its most-used internal platform at any one time in August, with every action screened before execution. Roughly one in 47,000 of more than 1 billion decisions was blocked. The company also said 6% of its AI research computing in a sample week in July went to safety research, rising to about 12% for research conducted by AI itself; the figures were conservative because dual-purpose work was classified as capability-related. The disclosure follows CEO Dario Amodei’s call for a slowdown in AI development and researcher Jacob Coxon’s resignation over concerns about the company’s approach. Separately, Nvidia was reported to be considering up to $10 billion for Anthropic’s potential IPO, which could seek as much as $100 billion at a valuation of about $2 trillion, while OpenAI was reportedly discussing funding at a valuation of around $1.2 trillion.