Anthropic’s Responsible Scaling Policy: Version 3.0

Anthropic's third version of its Responsible Scaling Policy (RSP) aims to address the evolving risks of AI systems by reinforcing effective measures, improving transparency, and introducing new measures for accountability. Initially introduced in September 2023, the RSP was designed to manage risks that could emerge rapidly with advancing AI technology, using a principle of conditional commitments tied to capability levels. While the RSP has successfully incentivized Anthropic and other companies to adopt stronger safeguards and inform early AI policy, challenges remain, particularly in achieving consensus on AI risks and implementing higher-level safeguards unilaterally. The updated RSP includes a clearer separation of company plans from industry recommendations, introduces a Frontier Safety Roadmap to outline risk mitigation strategies, and mandates regular Risk Reports with external reviews to enhance safety and transparency. As the AI landscape continues to evolve, Anthropic commits to revising its policy to adapt to new capabilities while encouraging multilateral action to address industry-wide risks.