Anthropic Updates Responsible Scaling Policy to Version 3.4

Anthropic released the first version of its Responsible Scaling Policy in September 2023. [1] The company notes that frontier AI models present new challenges and risks that warrant careful study and effective safeguards. [1] Anthropic has continued to refine the policy over time, expressing the belief that risk governance in this rapidly evolving domain should be proportional, iterative, and exportable. [1] The most recent iteration listed, Version 3.4, became effective on July 8, 2026. [1]

Anthropic's ongoing updates to its Responsible Scaling Policy reflect a steady focus on risk governance alongside model advancement. By consistently publishing policy iterations up to Version 3.4, the company maintains a highly visible framework for addressing the risks of frontier models, positioning its approach against competitors like OpenAI who also face intense scrutiny over safety practices and thresholds. The commitment to "iterative" governance suggests Anthropic expects rapid changes in capabilities that require regular regulatory recalibration. What the sources don't address: How the specific safeguards and mitigation requirements inside Version 3.4 differ practically from prior iterations.