Anthropic said Claude now “leads” 26% of the artificial intelligence research and development work inside the company, part of a new Anthropic R&D Automation Index it plans to publish regularly so outsiders can track how quickly AI is helping build the next generation of models. Reuters reported the figures Thursday from Anthropic’s company post.
As of August 2026, AI also collaborated with humans on more than 90% of the research work measured at the San Francisco startup. Claude is not operating fully autonomously in any part of the work covered by the index, Anthropic said. On the Epoch AI automation scale the company used, that share of work Claude “leads” has risen rapidly from about 1% in March.
The index catalogs AI R&D tasks at Anthropic, rates how automated each is, and aggregates the ratings. Epoch’s scale runs from no AI involvement through AI that collaborates, leads under human supervision, or operates fully autonomously. Anthropic framed the release as a public window into how frontier labs develop models amid concern that systems could eventually feed their own development with little human input.
On agent oversight, Anthropic said about 30,000 AI agents were doing research and engineering work on its main internal platform at any one time in August. Every action those agents take is screened before it is carried out; of more than a billion decisions that month, about one in 47,000 was blocked, the company and Reuters said.
On compute, Anthropic disclosed that about 6% of the computing power it used for AI research went to safety work in a sample week in July, rising to about 12% for research carried out by AI itself. The company called the figures conservative, counting dual-purpose safety-and-capability work as capability.
Anthropic said it plans to embed independent third-party evaluators with access comparable to internal risk teams to verify safety practices and monitor metrics such as these. The measurements sit alongside capability evaluations in the company’s Responsible Scaling Policy risk reports and its Advanced AI Framework transparency proposals.
This brief attributes the metrics to Reuters and Anthropic’s institute post. It does not invent additional automation levels or claim full autonomy.