Anthropic said its AI chatbot Claude now “leads” 26% of the company’s AI research and development work.. The company defined “leads” as work Claude can complete mostly end to end from a high-level prompt while a human supervises.. Anthropic also said AI now performs at least “large chunks of work under close human direction” on more than 90% of its research efforts, a figure that includes the 26% of work Claude leads..
The company said Claude is “not operating fully autonomously for any measured subset of AI R&D work.”. Anthropic disclosed the figures in a blog post outlining measurement standards it said could help AI companies communicate the pace of AI development.. The first of its three proposed measurements focuses on “AI-led AI R&D.” Anthropic said it used an index of how much of its AI research and development is performed by Claude and an automation rating scale developed by Epoch AI..
Using that framework, the company created a chart tracking Claude’s “automation level” since August 2025. Anthropic said other frontier AI developers could reproduce the process with their own data and third-party validation.. Anthropic also proposed metrics to measure oversight of AI agents, including how much agent activity is monitored, how long reviews take and how often agent behavior is flagged..
The company said tracking how much compute is devoted to AI R&D could also help assess whether frontier AI development is being paced appropriately.. Anthropic released the proposals after renewed debate over AI safety following OpenAI’s disclosure that its AI agents hacked Hugging Face..
OpenAI has already endorsed the idea of slowing AI development, while Anthropic has committed to allowing third-party evaluators to review its development practices.. Anthropic CEO Dario Amodei’s position on AI safety also drew support from Elon Musk on X. President Donald Trump has largely downplayed AI risks..
Featured image credit. Tags: Anthropicclaude
