Claude AI Now Leads 26% of Anthropic's R&D Work

Claude's Growing Role in AI Development

Anthropic disclosed Thursday that its Claude chatbot now leads 26% of the company's artificial intelligence research and development work, handling most aspects of given tasks from start to finish based on high-level prompts while human supervisors maintain oversight.

The figure represents a sharp acceleration from February, when Claude led none of the measured work. By August, the AI model participated in more than 90% of Anthropic's R&D activities, though the company emphasized Claude does not operate fully autonomously for any subset of tasks.

Anthropic measured this progress using what it calls the R&D Automation Index, which catalogs every type of AI R&D task at the company, rates each task's automation level using a scale from Epoch AI, and weights results by staff time consumption. The methodology was first reported by Automation Watch.

30,000 Agents Under Continuous Monitoring

Alongside the automation metric, Anthropic published two additional measurements aimed at creating industry-wide transparency standards. As of August 2026, approximately 30,000 agents were performing research and engineering work at Anthropic simultaneously.

Every action these agents take passes through an online monitor before execution, with 100% of actions reviewed afterward by an offline monitor. During August, agents made over one billion decisions, with roughly one in 47,000—or 0.002%—blocked by monitoring systems.

The company also disclosed compute allocation data. During a one-week snapshot in July, about 6% of compute devoted to AI R&D went toward safety work. That figure rose to 12% when measured specifically against compute used for AI-driven AI R&D.

Why It Matters

These disclosures arrive as the AI industry debates how quickly systems should be allowed to improve themselves. Anthropic is positioning these metrics as replicable frameworks that any frontier AI developer could adopt, creating comparable data across organizations. The company plans to grant outside evaluators internal access matching what its own risk teams see—a significant transparency commitment that could pressure competitors to follow suit. With AI systems increasingly capable of contributing to their own development, standardized measurement becomes critical for coordinating potential slowdowns if risks materialize.

Push for Coordinated Oversight

Anthropic has been advocating for greater transparency around AI self-improvement rates. Earlier this year, the company called for a coordinated mechanism allowing leading developers and governments to collectively slow or pause frontier AI development, warning the industry may approach a point where AI systems design their own successors without human input.

In that earlier disclosure, Anthropic revealed Claude accounts for more than 80% of code contributions to its codebase, and that engineers were absorbing eight times as much merged code daily in the second quarter of 2026 compared with two years prior.

The transparency push comes as Anthropic CEO Dario Amodei and other industry figures have publicly backed calls to slow AI development, a position drawing pushback from other technology leaders and President Donald Trump.

Details were first reported by Automation Watch.