Anthropic says Claude leads 26% of its R&D
Anthropic has published internal metrics on how much of its AI research Claude now performs, a rare frontier-lab disclosure meant to give outsiders visibility into how close labs are to self-improving AI.
Anthropic published a set of proposed measurements for tracking the pace of AI development inside frontier labs, along with a snapshot of its own figures. The metrics cover three areas: how much AI is building its own successors, how well humans can oversee AI agents on internal systems, and the resources behind more capable models.
The headline number comes from a prototype Anthropic R&D Automation Index. It catalogues every kind of AI R&D task at the company and rates each one on an automation scale developed by Epoch AI. The scale runs from AL0 (no AI involvement) to AL5 (fully autonomous). As of August 2026, Anthropic reports:
- Claude "leads" 26% of its AI R&D work (AL4: it completes most of a task end-to-end from a high-level prompt while a human supervises).
- More than 90% of the work is at or above "AI collaborates" (AL3).
- Claude does not operate fully autonomously on any measured subset of the work.
According to the material provided, the post also reports that about 30,000 agents run at once on Anthropic's main internal platform and that 6% of R&D compute goes to safety work. The source excerpt is cut off before the full oversight and compute sections.
Anthropic acknowledges two limits on its numbers. They are self-reported, and Claude is partly used to judge Claude, so the evaluator could make the same errors as the system it checks. There is also no shared methodology across labs. The company says it plans to embed independent third-party evaluators from multiple organizations, with access to internal systems and data, to verify its safety practices and monitor metrics like these.