Anthropic published three internal measures of the pace of AI development inside its own labs and called on other frontier labs to disclose the same indicators. As of August 2026, on Epoch AI's automation scale, Claude did not fully autonomously run any measured AI R&D work, but it 'led' 26% of Anthropic's AI R&D work, and AI collaboration or above accounted for more than 90%, up from less than 1% AI-led in February 2026. Anthropic says about 30,000 agents work on R&D and engineering on its most-used internal platform, 100% of agent actions pass an online monitor before execution, and roughly 100,000 records are flagged offline each week, with about 50 escalated to human review. For July 13-20, 2026, about 6% of AI R&D compute was allocated to safety research.

Editor's note

This summary is based on the primary source. Open the original for full context, data, and qualifying details.