CONCEPT
AI Safety Levels (ASL)
The tiered classification at the heart of the Responsible Scaling Policy — capability thresholds analogous to biosafety levels, specifying the safety measures required before deployment at each level.
AI Safety Levels (ASL) are the tiered classifications at the heart of Anthropic's Responsible Scaling Policy, defining capability thresholds that trigger specific safety requirements before deployment. Modeled on the biosafety levels used for working with pathogens of varying danger, the ASL framework specifies what safety measures must be in place before a system exhibiting a given level of capability can be deployed at scale. ASL-1 covers systems posing no meaningful catastrophic risk. ASL-2 covers systems with early signs of dangerous capabilities that do not yet provide meaningful uplift to bad actors. ASL-3 covers systems that substantially increase the risk of catastrophic misuse or show early signs of autonomous capabilities. ASL-4 and above address capabilities the current framework treats as requiring additional research before commitments can be made.
In The You On AI Field Guide
The biosafety analogy is deliberate and illuminating. In biological research, the containment level required for working with a pathogen is determined by the pathogen's characteristics —