Anthropic Reveals Claude Leads 26% of Internal AI Research Tasks

Anthropic introduced an R&D automation measurement framework and disclosed internal data showing that Claude leads 26% of the company's AI research and development tasks as of August 2026, up from less than 1% in February 2026. Additionally, over 90% of internal AI R&D tasks involve Claude at a collaborative level or higher, though zero tasks have achieved complete autonomy (AL5). This benchmark provides rare visibility into how frontier AI labs are utilizing their own models to accelerate AI development, tracking progress toward recursive self-improvement. By sharing these metrics publicly, Anthropic aims to establish transparent standards for evaluating AI capability growth, compute allocation, and safety governance across the industry. The index adapted Epoch AI's six-tier automation scale (AL0–AL5) and evaluated 15,000 internal task samples categorized into 542 task nodes using an independent Claude model judge. Anthropic also revealed that 6% of its total AI R&D compute (and 12% of AI-driven R&D compute) is dedicated to safety research, while a real-time monitoring system intercepts roughly 0.002% of risky agent actions.