Anthropic says its Claude AI is now helping build the company’s next, more capable generation of models, marking a notable step in the use of artificial intelligence to advance AI research itself.
Claude currently leads 26% of Anthropic’s model research and development work, according to a company announcement Thursday. Anthropic defines that level of involvement as the ability to handle most of a task from a high-level prompt through completion, while people continue to supervise the process. The system is not operating fully autonomously.
More broadly, Claude is involved in roughly 90% of the company’s research and development through what Anthropic describes as collaboration. In those cases, the AI can complete substantial portions of work but remains under close human direction.
The disclosure arrives as prominent voices in artificial intelligence, including Anthropic CEO Dario Amodei, increasingly call for the industry to slow the pace of development because of safety concerns.
As policymakers and technology leaders debate how quickly AI should advance, Anthropic argued that the public should have greater visibility into the capabilities being developed inside frontier AI labs. “We should do everything possible to minimize the gap between what frontier labs know and what the public knows,” the company wrote in a blog post. That includes more rigorous measurement, public reporting and an opportunity for society to determine how the information should be used.
Anthropic warned that AI models capable of speeding up their own development may become harder for humans to understand and control. Publishing these measurements, it said, could improve awareness of how close major AI companies may be to recursive self-improvement — the point at which a model can autonomously create its own successor.
MORE STORY: Mets’ Jared Young Stretchered Off After Collision With Nolan McLean
The company called on other AI developers to release comparable metrics regularly and to adopt a public methodology, allowing progress to be tracked over time and compared across different labs.
Anthropic’s update did not make clear how near it believes it is to recursive self-improvement. Still, Claude’s rapidly expanding role in research and development stands out. In February, the model was leading none of the company’s R&D work. By August, six months later, it had reached the point of leading about one-quarter of that work while under human supervision.
Anthropic also outlined its oversight practices for AI agents, saying that about 30,000 agents were conducting research and engineering tasks as of August. The company said these safeguards are important for measuring how often monitoring systems catch agent misbehavior. Anthropic recently pledged to bring in independent third-party evaluators who will be embedded within the company to assess its safety work.
The latest debate over AI safety intensified after an Anthropic researcher resigned last week, issuing a stark warning about the technology’s potential risks to humanity. Amodei, OpenAI CEO Sam Altman, Elon Musk and other technology figures have backed calls to slow development. Others in the industry, along with President Donald Trump, have pushed against that approach.
Copyright 2026 The Associated Press. All rights reserved. This material may not be published, broadcast, rewritten or redistributed without permission.