Anthropic, a leading artificial intelligence research company, has revealed that its large language model, Claude, is now independently leading a significant portion of its internal AI research and development (AI R&D) work. This rapid acceleration, with Claude leading 26% of tasks by August 2026, marks a substantial jump from less than 1% earlier in the year.
Measuring AI's Role in Development
To track this evolving dynamic, Anthropic developed an "R&D Automation Index." This index categorizes AI involvement from AL0 (no AI) to AL5 (full autonomy). The "AL4" level, where an AI "leads," means it can complete most of a task from a high-level prompt with human supervision. While Claude has reached this AL4 level for a quarter of Anthropic's measured R&D, the company emphasizes that it has not achieved full autonomy in any area.
The distinction is important: Anthropic is not claiming Claude can independently build its successor. Human involvement and oversight remain fundamental to the process.
The Shift Towards Recursive AI Development
The growing role of AI in developing other AI systems is a critical indicator for the future of the field. Anthropic is particularly interested in "recursive self-improvement," a point at which AI systems could independently contribute to creating more capable successors. Although this ultimate goal has not yet been reached, the increasing automation of research, coding, testing, and engineering tasks by AI agents could dramatically shorten development cycles for frontier models, accelerating overall AI progress.
Scale of AI Agent Operations
Beyond specific R&D leadership, Anthropic also employs a vast network of AI agents across its research and engineering operations. As of August 2026, approximately 30,000 such agents were simultaneously active on the company's primary internal platform. These agents' actions are rigorously monitored, with both online and offline systems reviewing their decisions. Out of over a billion decisions made in August, a tiny fraction (0.002%) were blocked by the online monitor, with slightly more flagged for human review by the offline system.
Investment in AI Safety
Alongside its development efforts, Anthropic is also measuring its commitment to AI safety. During one week in July 2026, roughly 6% of the total computing resources dedicated to AI R&D were allocated to safety work. For compute specifically used in AI-driven AI R&D, this figure rose to about 12%. Anthropic notes these are conservative estimates, acknowledging that human expertise is also a significant, though harder to quantify, safety investment.
Call for Industry Transparency
Anthropic advocates for greater transparency across the frontier AI industry. The company urges other AI labs to publish similar, comparable metrics regularly. This would allow for better tracking of AI development pace and foster a more informed public discourse, despite acknowledged challenges in standardizing methodologies and the complexities of AI evaluating AI-generated work.