Anthropic’s Responsible Scaling Policy (RSP) establishes AI Safety Levels (ASL)

AI Impact Summary

Anthropic is implementing a tiered safety framework, the AI Safety Levels (ASL), to manage the escalating risks associated with increasingly capable AI models like Claude. The policy establishes ASL-1 through ASL-5+ levels, with ASL-2 currently representing Anthropic’s current safety and security standards for Claude. This framework necessitates a ‘race to the top’ dynamic, incentivizing rapid safety advancements while acknowledging the potential for temporary pauses in model training to ensure compliance with stricter ASL requirements, mirroring pre-market testing in industries like automotive and aviation.