Anthropic Says Claude Now Leads a Quarter of Its Model R&D
Anthropic says Claude leads 26% of its model research and development and is involved in about 90% of R&D work, as the lab calls for public reporting on AI progress.
Anthropic has disclosed that its AI model Claude is playing a growing role in developing the next, more capable version of itself, with the company saying the system now leads 26% of its model research and development work.
The lab described that level of involvement as the model completing most of a task end-to-end from a high-level prompt, though still under human supervision. Claude is not yet operating fully autonomously, the company said.
Beyond the work it leads, roughly 90% of Anthropic's research and development is carried out in collaboration with Claude, which the company characterised as the model handling large chunks of work under close human direction.
The shift has been rapid. In February, Claude led none of the company's research and development. Six months later, it was leading a quarter of that work, a benchmark Anthropic said was reached in August.
The disclosure came as several leading figures in artificial intelligence, including Anthropic chief executive Dario Amodei, have called for a slowdown in the technology's development over safety concerns.
In a blog post, the company argued that as leaders weigh how quickly to pace AI development, the gap between what frontier labs know and what the public knows should be minimised. It said this means measuring AI development better, reporting on it publicly, and giving society a chance to decide how to use that information.
Anthropic warned that models accelerating their own development could make it harder for humans to understand or control such systems. It said sharing these metrics could improve understanding of how close leading labs are to recursive self-improvement, the ability of a model to autonomously build its successor.
The company urged other AI developers to publish similar metrics regularly, encouraging the use of a public methodology so the figures could be compared over time and potentially across labs. It was not clear from the disclosure how close Anthropic believes it is to achieving recursive self-improvement.
Anthropic also shared details of its oversight measures for AI agents, noting that about 30,000 agents were doing research and engineering work as of August. Such oversight is important for tracking how often agent misbehaviour is caught by monitoring systems, the company said. It recently committed to embedding external third-party evaluators within the company to monitor safety efforts.
The debate over AI safety has intensified after an Anthropic researcher resigned last week with a warning about the threats the technology poses to humanity. Amodei, OpenAI chief executive Sam Altman, Elon Musk and other tech leaders have since backed the idea of slowing development, while other technology figures and President Donald Trump have pushed back.