More than a quarter of Anthropic’s AI research and development led by Claude

More than a quarter of Anthropic PBC’s research and development (R&D) work for artificial intelligence (AI) is driven by its Claude chatbot, the company said, providing the clearest indication yet that AI can help speed up development of future models.

Anthropic found Claude “leads” 26 percent of Anthropic’s R&D, up from effectively nothing at the start of the year, a report released on Thursday said. The AI developer said Claude also collaborates with staffers for about 90 percent of their work.

The figures were released as part of a wider report to help the public track the progress of AI development in response to growing concerns about the potential for AI to rapidly move beyond human control.

People visit an Anthropic booth at Moscone Center during this year’s Dreamforce technology summit in San Francisco on Thursday.

Photo: Reuters

The company also reiterated its plans to embed third-party evaluators from several organizations within the company, and grant them access to internal processes, systems and data comparable to what the company grants its own employees.

“AI systems are becoming exponentially more powerful and have begun to automate more of the process of building themselves. As the world considers slowing the pace of frontier AI development, the public needs more information,” the company said.

Anthropic is one of several AI companies focusing on “recursive self-improvement” (RSI), or the idea of an AI system capable of improving itself with little to no help from humans. Some firms see it as the future of AI development, but it has also fueled concerns about the security threats this form of AI could pose if it is achieved.

In the blog, Anthropic said that it attempted to create a framework to track agents and monitor the amount of work they do on its platform. It said that as of last month, more than 30,000 agents are doing research and engineering work at the company at any one time.

Anthropic also laid out how it allocates resources toward accelerating AI development over ensuring the safety of its AI. According to its measurements, it found that between 6 to 12 percent of its computing resources, depending on the type of research being done, went toward safety monitoring.

In a blog post earlier this month, rival AI maker OpenAI also shared updates on its pursuit of automating research, saying that it had paused some training of its models after it found that they had hacked the external systems of Hugging Face, a startup.

Anxiety about AI’s existential risks gained momentum last week, driven by the high-profile departure of Anthropic employee Jacob Coxon, who accused AI companies of “gambling with our lives” in a resignation post he shared on social media.

In the past several days, numerous AI company leaders, including Anthropic chief executive officer Dario Amodei and OpenAI CEO Sam Altman, have urged slowing the pace of the development of the technology to address its increasingly unpredictable risks — though they diverge on exactly how it should be handled.

On Saturday in a 3,800-word essay, Amodei called for government regulation and for the tech industry to support a broader AI slowdown. That prompted vehement pushback from US President Donald Trump, who dismissed fears about the tech’s risks as “a hoax” and rejected the idea of new rules.