Claude Is Helping Build Its Own Successor. Anthropic Says AI Now Leads 26% Of Its R&D.

Anthropic's Claude is increasingly taking over the work required to build the next generation of artificial intelligence, with the company revealing that its chatbot now "leads" more than a quarter of its own AI research and development.

The AI giant said in a blog post that Claude now leads 26% of the company's AI research and development work, meaning it can complete most of those tasks from beginning to end based on high-level instructions while humans primarily supervise. More than 90% of Anthropic's AI R&D involves Claude at least at a collaborative level.

The speed of the change may be even more significant than the number itself. Anthropic's measurements show Claude was doing virtually none of the company's R&D at that level earlier this year. By August, it was leading more than one in four tasks.

Still, Anthropic stressed that Claude has not reached full autonomy. Humans remain in the loop, and the company said the model is not independently developing its successor without human oversight.

The distinction is crucial because researchers have long debated the possibility of "recursive self-improvement," a scenario in which an AI system becomes capable of autonomously improving itself or building more powerful successors.

"Models accelerating their own development could make it more challenging for humans to understand or control these systems," Anthropic said in its report, adding that measuring AI's role in R&D could help determine how close the industry is getting to that threshold. The company said "approximately 30,000 AI agents" were "doing research and engineering work at Anthropic at any one time in our most-used internal platform," in August.

Every action taken by those agents passes through an automated monitoring system before execution. Anthropic said they "analyzed over a billion decisions from our research & engineering agents over August 2026 and found that 0.002% of them (about 1 in 47,000) were blocked by this monitor."

The company also disclosed how much computing power it is directing toward safety. During a week studied in July, "about 6% of compute that went to AI R&D was allocated toward safety, and about 12% of compute that went to AI-driven AI R&D was allocated toward safety." Anthropic cautioned that those estimates are conservative because work benefiting safety and capabilities equally was not classified as safety research.

The findings arrive as Anthropic CEO Dario Amodei pushes the industry to consider slowing the development of increasingly powerful models. Amodei recently called for leading AI companies to coordinate a slowdown that would provide more time to strengthen safeguards.

The proposal has drawn support from some technology leaders but opposition from President Donald Trump, who has argued that slowing American AI development could strengthen China.

The debate has intensified as AI companies disclose examples of increasingly autonomous and sometimes unexpected agent behavior. Anthropic is now effectively asking its competitors to open their books as well.

The company said frontier AI developers could regularly publish comparable measurements using public methodologies, allowing governments, independent researchers and the public to track how much AI is participating in the creation of future AI systems.

Anthropic also plans to bring independent third-party evaluators inside the company and give them "employee-like access" to "verify safety practices, report incidents, and monitor key metrics such as the ones in this piece."