AI Safety Levels

A tiered classification system, used in Anthropic's Responsible Scaling Policy, that assigns a model a safety level based on the catastrophic risks its capabilities could pose, with each higher level requiring stronger security, deployment and containment safeguards before release.