Anthropic Lets Claude Lead Its Own Research

🔥 What happened Anthropic just published internal numbers on how much Claude automates its own AI research. As of August 2026, Claude "leads" 26% of Anthropic's AI R&D work end-to-end, and over 90% runs at least in collaboration mode. No measured area is fully autonomous yet. 💡 Why it matters The Anthropic R&D Automation Index uses Epoch AI's AL0–AL5 scale — at AL4, the AI completes a task from a high-level prompt while a human supervises. 26% at that level means a quarter of the research building the next model is already done by the current one. Recursive self-improvement just moved from thought experiment to dashboard metric. ⚡ Our take A lab reporting 26% self-automation while publicly calling for a slowdown either has a conscience or a very good comms team. Either way, watch the number, not the press release.