Definition
OpenAI’s internal framework for evaluating and mitigating catastrophic risks from frontier AI models before and after deployment.
Key Points
- Covers capability thresholds, red-teaming, and post-deployment monitoring
- Cited by critics as insufficient without broader safety-culture change