Anthropic reports Claude drives 26% of its own R&D work

Sep 17, 2026

Anthropic PBC reports that its Claude chatbot now leads 26% of its internal AI research and development, up from 1% in March 2026. The startup also reveals that Claude collaborates on over 90% of its R&D work. To increase transparency and manage existential risks, Anthropic has proposed three public metrics to track AI progress: AI-led R&D, agent autonomy, and compute allocation, disclosing that it currently runs 30,000 autonomous agents and dedicates up to 12% of its AI-driven compute to safety.

Claude AI self-development metrics

▪Claude's contribution to Anthropic PBC's development rose from 1% in March 2026 to 26% in August 2026, based on a scale developed by Epoch AI
▪Anthropic PBC reported that its Claude chatbot leads 26% of the company's artificial intelligence research and development work
▪Anthropic PBC stated that Claude collaborated with humans on more than 90% of the company's research and development work as of August 2026

AI development slowdown coordination

▪Anthropic PBC researcher Evan Hubinger warned that rapid AI advancement poses a greater than 10% chance of killing all humans within the next 10 years
▪Anthropic PBC Chief Executive Dario Amodei published a three-step plan calling on leading model makers to coordinate on slowing down the pace of frontier AI model development

AI agent autonomy monitoring

▪Anthropic PBC screened over one billion decisions made by its internal AI agents in August 2026, blocking approximately one in 47,000 actions
▪Anthropic PBC reported running approximately 30,000 autonomous AI agents on its main internal platform at any one time in August 2026

Compute allocation tracking

▪Anthropic PBC proposed tracking compute allocation as a verifiable input to monitor where AI developers focus their efforts over time
▪Anthropic PBC took a compute snapshot from July 13 to July 20, 2026, to establish a baseline of how its processing resources are distributed

AI safety research resources

▪Anthropic PBC allocated 12% of its computing power dedicated to AI-driven research and development toward safety-focused tasks in July 2026
▪Anthropic PBC allocated 6% of its total computing power used for AI research to safety work during a sample week in July 2026