Anthropic Revises RSP Safety Policy as AI Competition Intensifies

Anthropic introduced its Responsible Scaling Policy (RSP) in 2023, using AI Safety Level capability thresholds to constrain frontier models. In principle, the company would pause further scaling or deployment if training progress outpaced existing safeguards. The voluntary framework was regarded as one of the AI industry's clearest safety commitments and influenced OpenAI and Google DeepMind to establish similar systems.

On February 24, 2026, Anthropic released RSP 3.0, removing its firm commitment to halt training when risk mitigations could not be assured in advance. It instead adopted a Frontier Safety Roadmap and pledged to publish a Risk Report every three to six months. The company raised $30 billion that month at a valuation of $380 billion. The new policy received unanimous backing from CEO Dario Amodei and the board.