Anthropic just published the internal metrics it uses to measure how fast frontier AI is advancing, pulling back the curtain on a question the industry has argued about for two years. The company says its own models now lead 26% of AI R&D work internally, a number it frames as proof that self-improvement is no longer theoretical. The disclosure lands days after CEO Dario Amodei publicly called for the pace of frontier development to slow down.
⚡ Fast Takeaways:
- Core Update: Anthropic detailed three practical metrics for monitoring development speed, marking the first time it has shared these numbers outside the lab.
- Key Metrics / Specs: Claude reportedly handles 26% of internal AI R&D work, with Anthropic explicitly flagging self-improvement as a risk worth tracking.
- Access & Availability: The metrics framework is public now, aimed at policymakers, rival labs, and enterprises weighing oversight rules.
- Context: The release follows rising "rogue agent" incidents and loud industry disagreement over whether frontier work should decelerate.