Anthropics Responsible Scaling Policy

Anthropic’s original Responsible Scaling Policy, announced in September 2023, tied stronger AI capabilities to stricter safeguards. Its AI Safety Levels described escalating requirements for security, testing and deployment as models became more capable of causing catastrophic harm. The company said this could require pausing further training when its safety procedures could not keep pace. The original announcement made safeguards a condition for further scaling.

That is a historical commitment. A later revision, announced in February 2026, separated mitigations the company says it will pursue itself from broader recommendations for industry, and added a nonbinding public roadmap of safety goals plus periodic risk reports. The revision announcement explains the changed framework.