#ai-safety
1 articles
- 2026-10-21
Who Watches the AI Watcher? A Monitor Score Is Not an Outcome
New work from METR and OpenAI shows why a model-monitor score cannot substitute for end-to-end outcomes, retry accounting, and independent scrutiny.
1 articles
New work from METR and OpenAI shows why a model-monitor score cannot substitute for end-to-end outcomes, retry accounting, and independent scrutiny.