METR reviews Anthropic's automated R&D risk claims
The key detail is that METR worked from the nonpublic version of the report, meaning its findings may differ from what Anthropic published openly. Lab-commissioned external evaluations of capability thresholds are becoming a reference point for safety commitments, so how METR judges Anthropic's claims sets a precedent other labs may face. Readers tracking AI safety governance and frontier-lab accountability should follow the full assessment.
