At Anthropic, Claude is playing an increasingly prominent role in the development of next-generation artificial intelligence, with the company’s chatbot now assuming a leading position in over a quarter of its AI research and development endeavors, where it “leads” the work.
Evolution of Internal Workflows
According to Anthropic, Claude is currently at the helm of 26% of the company’s AI research and development initiatives, enabling it to handle the majority of tasks autonomously, from start to finish, based on broad guidelines, while human oversight is primarily focused on supervision. Furthermore, Claude is involved, to some extent, in more than 90% of Anthropic’s AI R&D activities.
The rapid pace at which Claude has taken on this role is noteworthy, as earlier this year, it was not contributing to the company’s R&D efforts at this level. However, by August, Claude had become a leader in over one-quarter of the tasks. Despite this, Anthropic emphasizes that Claude has not yet achieved complete autonomy, as human intervention remains essential, and the company asserts that Claude is not independently developing its successor without human oversight.
Monitoring Autonomous Decision-Making
For some time, researchers have been exploring the concept of “recursive self-improvement,” a scenario where an AI system becomes capable of autonomously enhancing itself or creating more powerful successors. In August, Anthropic reported that “approximately 30,000 AI agents” were engaged in research and engineering work on its internal platform, with every action taken by these agents subject to an automated monitoring system prior to execution.
Anthropic analyzed over a billion decisions made by its research and engineering agents in August 2026 and found that 0.002% of them (approximately 1 in 47,000) were blocked by this monitoring system.
Emphasis on Safety and Transparency
The company has also disclosed the extent to which it is allocating computing resources toward safety. In a week-long study conducted in July, Anthropic found that “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.” The company notes that these estimates may be conservative, as work that benefits both safety and capabilities equally was not classified as safety research.
These findings come as Anthropic CEO Dario Amodei advocates for the AI industry to consider slowing the development of increasingly powerful models, allowing more time to strengthen safeguards. Amodei’s proposal has garnered support from some technology leaders but has been met with opposition from President Donald Trump, who argues that slowing American AI development could strengthen China.