Anthropic publishes the numbers behind "pace": Claude now leads 26% of its own R&D

Anthropic published three transparency metrics on September 17. Its argument: if the industry is going to debate "pacing the frontier," the idea CEO Dario Amodei floated days earlier that labs should deliberately slow capability gains, the public needs numbers to debate with. "As the world considers pacing the frontier, we should do everything possible to minimize the gap between what frontier labs know and what the public knows," the company wrote, and it disclosed its own internal figures to back that up.

The three metrics

Automation: Claude now leads roughly 26% of Anthropic's own AI research and development work, up from under 1% six months earlier. The scale runs 0-5, with "leads" (AL4) above "collaborates" (AL3), and over 90% of the company's R&D work now clears the "collaborates" bar or higher. Anthropic notes Claude isn't "operating fully autonomously" for any measured subset of that work.

Oversight: roughly 30,000 agents run research and engineering work on Anthropic's internal platform at any given time, with real-time and after-the-fact monitor coverage. The block rate on flagged decisions is about 0.002%, roughly 1 in 47,000.

Compute: in a July 13-20 snapshot, about 6% of AI R&D compute went to safety work.

Numbers others can check

Anthropic is asking OpenAI, Google DeepMind and every other frontier lab to publish the same three categories on a recurring basis, with independent third-party evaluators eventually verifying them. That would let customers, regulators and competitors track safety claims quarter over quarter. A self-reported automation share is still self-reported. It is a better starting point than no number, and it sets the bar the next lab's disclosures will be measured against.