Anthropic reveals Claude 'leads' 26 percent of its internal AI research work
Anthropic has published new metrics showing how much of its own AI research and development is handled by its chatbot Claude, saying it "leads" 26 percent of that work as part of a broader proposal for measuring AI development pace industry-wide.
Anthropic has released new figures detailing how much of its own artificial intelligence research and development work is carried out by its chatbot Claude. In a blog post proposing standards for measuring the pace of AI progress, the company stated that Claude "leads" 26 percent of its AI R&D work.
According to Anthropic, "leads" means the AI can complete most of a task from start to finish based on a high-level prompt, while a human supervises the process. The company further claims that more than 90 percent of its research involves AI doing at least large portions of the work under close human direction — a figure that includes the 26 percent Claude leads outright.
This suggests Claude is involved in the majority of research work done at Anthropic, though the company stresses the chatbot is not operating fully autonomously in any measured category.
A new way to track AI progress
These figures come from the first of three measurement approaches Anthropic is proposing, focused on "AI-led AI R&D." Using an index of how much of its research and development is performed by Claude, combined with an automation rating scale developed by Epoch AI, Anthropic built a chart tracking Claude's "automation level" since August 2025. The company believes any leading AI developer could replicate this approach using its own data, validated by an independent third party.
Anthropic also proposed methods for measuring oversight of AI agents — including how much of their activity is monitored, how long reviews take, and how often problematic behavior is flagged — as well as tracking how much computing power is devoted to AI research.
The goal, the company says, is greater transparency that could help the industry respond if AI development starts to move out of control, or at minimum give the public a clearer picture of what is happening. So far, getting companies to agree on actually slowing development down remains the harder challenge, even as some, including Anthropic, commit to third-party review of their practices.
