Anthropic Says Claude Leads 26 Percent of Its AI R&D Work

Anthropic has shared a striking figure about how much of its own research its AI does. In a new blog post laying out measurement standards it believes could help AI companies communicate the pace of development, the company said its chatbot Claude "leads" 26 percent of its AI R&D work.

By "leads," Anthropic means Claude can complete most of a task end-to-end from a high-level prompt while a human supervises. The company also claims AI now handles at least large chunks of work under close human direction on more than 90 percent of its research, a figure that includes the 26 percent Claude leads. Still, Anthropic says the chatbot is "not operating fully autonomously for any measured subset of AI R&D work."

The numbers come from the first of three proposed measurements, focused on "AI-led AI R&D." Using an index of how much of its research Claude performs and an automation rating scale developed by Epoch AI, Anthropic charted Claude's "automation level" since August 2025 - a process it believes any frontier model maker could reproduce with its own data and third-party validation.

Anthropic also proposed ways to measure oversight of AI agents - how much activity is monitored, how long review takes and how often behavior is flagged - and to track compute devoted to AI R&D. The hope is that more transparency could make it easier to respond to out-of-control development, or at least give the public a clearer view.

Getting companies to agree on slowing AI development may not be the hard part. OpenAI has already paid lip service to the idea, and Anthropic has committed to letting third-party evaluators review its practices. Whether the industry faces anything beyond self-regulation is far less certain - President Donald Trump has largely downplayed AI's risks.