Claude now leads 26% of Anthropic’s own development
Claude now leads 26% of Anthropic’s own development and contributes to roughly 90% of the lab’s research and development work, the company said Thursday in an announcement on recursive self-improvement.
The figure of 26% means Claude can complete most of a given task “end-to-end from a high-level prompt” while remaining under human supervision, Anthropic said. The model is not working completely autonomously.
The 90% figure refers to work Claude does in “collaboration” with humans, which the company defined as the model handling “large chunks of work under close human direction.”
Anthropic said models that accelerate their own development could make it “more challenging for humans to understand or control these systems.” The company shared the metrics, it said, to help society gauge how close leading AI labs are to recursive self-improvement, or a model’s ability to autonomously build its successor.
“We should do everything possible to minimize the gap between what frontier labs know and what the public knows,” Anthropic said in a blog post. “This means better measuring the development of AI, reporting on it publicly, and giving society an opportunity to decide how to use this information.”
The company urged other AI developers to share similar metrics on a regular basis, using a shared public methodology so numbers could be compared over time and across labs.
The portion of research and development Claude “leads” was zero in February, Anthropic said, and rose to 26% by August.
It was unclear from the disclosure how close Anthropic believes it is to recursive self-improvement.
Anthropic also shared details of its agent oversight measures, noting that approximately 30,000 agents were doing research and engineering work as of August. The company said oversight measures are crucial for tracking how often agent misbehavior is detected by monitoring systems.
Anthropic recently committed to setting up external third-party evaluators embedded within the company to monitor safety efforts.
The announcement came as some leading figures in AI, including Anthropic CEO Dario Amodei, have called for a slowdown in the technology’s development over safety concerns.
An Anthropic researcher kicked off much of the recent AI safety dialogue when he resigned last week with a warning about the threats the technology poses to humanity, according to Anthropic. Amodei, OpenAI CEO Sam Altman and Elon Musk have since supported the idea of slowing development, while other tech leaders and President Donald Trump have pushed back.
