XNews

Anthropic Publishes Internal AI R&D Dashboard Metrics

Anthropic Publishes Internal AI R&D Dashboard Metrics

Anthropic publicly shared its internal AI R&D dashboard, revealing that as of July 2026, Claude dominates 26% of AI R&D work (AL4 level), with over 90% of R&D involving AI deeply. The dashboard also shows 30,000 agents running simultaneously, with 6% of compute allocated to safety, and a monitoring system intercepting 0.002% of agent actions.

Key insights

R&D Automation Index: 26% at AI-Dominated Level

Anthropic measured that 26% of R&D work has reached AL4 (AI-dominated), where humans give high-level instructions and AI completes most work end-to-end. Over 90% of work is at AL3 or above. No work subset has reached AL5 (fully autonomous). This baseline provides a direct window into the acceleration of recursive self-improvement.

Agent Oversight: 30,000 Agents with Two-Layer Monitoring

At any time, about 30,000 agents are working on R&D and engineering. Online monitoring intercepts 0.002% of actions (about 1 in 47,000), typically fast and irreversible harms like exfiltration of model weights. Offline monitoring flags about 100,000 conversation records per week, with about 50 escalated to human review. The system is only a few months old, and its effectiveness is still being evaluated.

Compute Allocation: 6% for Safety Work

Analysis of one week of compute tasks showed that about 6% of compute used for AI R&D went to safety work. When looking only at AI-driven AI R&D, the safety share was about 12%. These are conservative estimates. The methodology allows for cross-lab comparisons and potential future commitments to safety compute ratios.

Source posts