Anthropic reported that its Claude models now "lead" 26% of the company's AI research and development work as of August 2026, up from under 1% in February — the first results from a new R&D Automation Index published September 17.
The index, built on an Automation Level scale (AL0-AL5) developed by Epoch AI, sampled roughly 15,000 tasks from 20% of staff across departments in July 2026, sorted into 542 categories by Claude itself. Work at the "leads" level (AL4) means Claude carries most of a task from a high-level prompt while a human supervises; more than 90% of measured R&D work sits at "collaborates" (AL3) or higher, but none reached full autonomy (AL5).
Anthropic said around 30,000 AI agents operate on its internal platform at any given time, with automated monitors blocking just 0.002% of more than a billion decisions, and 6% of AI R&D compute was set aside for safety work during a sample week in July.