Anthropic Responsible Scaling Policy Version 3.0

Establishes a framework for managing catastrophic risks from advanced AI, focusing on mitigating risks through industry-wide safety measures. Requires the company to develop Frontier Safety Roadmaps detailing security, alignment, safeguards, and policy goals. Mandates Risk Reports that evaluate and disclose AI model risks and risk mitigation effectiveness, subject to internal and external review. Encourages industry-wide safety by setting capability thresholds that trigger specific mitigations and recommend governance by third parties or independent organizations. Highlights risk mitigation strategies against threats such as chemical/biological weapons production, sabotage opportunities, and automated R&D. Outlines governance commitments, including appointing a Responsible Scaling Officer and enabling internal transparency and accountability measures. Details the company's approach to balancing safety with competitive pressures and outlines procedural updates for risk management and reporting, offering strategic guidance for AI development and deployment practices.