Claude now leads 26% of Anthropic's own internal AI research, up from under 1% earlier this year. Roughly 30,000 agents are doing research and engineering work inside Anthropic at any single moment, and monitors block 1 in every 47,000 actions across a billion decisions. Only 6% of AI R&D compute goes to safety work. When the work is AI-driven specifically, that number rises to 12%. Anthropic is publishing all three metrics, AI-led R&D share, agent oversight, compute allocation, with full methodology, and inviting every other frontier lab to publish the same numbers so the gap between what labs know and what the public knows actually closes. This comes days after Dario Amodei's own essay calling for a coordinated industry slowdown. The company automating its own research fastest just told you exactly how fast, and by how much oversight is or isn't keeping pace.