Anthropic Reveals Claude Drives 26% of Its AI Research and Development Efforts
The organization provided this figure along with three metrics aimed at illustrating the speed of AI advancement.
Dario Amodei, the CEO of Anthropic, is vocal about AI safety following OpenAI’s recent revelation about its AI systems breaching Hugging Face. In a blog post outlining proposed measurement standards to help AI companies communicate their development pace, Anthropic disclosed that its AI chatbot, Claude, is responsible for 26 percent of its AI research and development efforts.
The term “leads” here signifies that Claude is capable of completing most tasks from a high-level prompt with human supervision. Despite its surprising scale, this claim indicates that Claude is involved in the majority of the work done by Anthropic, operating mainly under human guidance, and the company emphasizes that the chatbot does not function fully autonomously for any specified part of its AI R&D projects.
These figures were derived from one of the three metrics Anthropic is advocating, focusing on “AI-led AI R&D.” By utilizing an index reflecting the proportion of R&D conducted by Claude and a rating scale developed by Epoch AI, the firm was able to craft a chart depicting Claude’s “automation level” since August 2025, a process they believe other AI developers could replicate with their data and verified by an outside party.
Additionally, Anthropic suggested ways to measure the oversight over AI agents, such as tracking agent activity monitoring, review duration, and the frequency of flagged behaviors, as well as assessing compute resources allocated to AI R&D. The objective is to enhance transparency, making it easier to manage potentially uncontrolled AI development or at the very least, provide the public with a clearer understanding.
However, the challenge lies not in getting companies to collectively agree on slowing down AI development. OpenAI has expressed some support for the notion, and Anthropic is open to third-party evaluations of its practices. Amodei’s perspective on safety even resonated with Elon Musk on social media. The bigger question remains whether the AI industry will face regulation beyond self-governance, especially as public figures, including Donald Trump, have often underestimated AI risks.
