Anthropic Plans Data Retention Policy Change

Anthropic has announced that all prompts and outputs from its most capable Mythos and Fable models will be logged for 30 days, with no opt-out, across both first-party and third-party platforms, a policy designed for safety monitoring rather than training. The company emphasizes that retained data will be used to detect new threats and vulnerabilities, and will be automatically deleted after 30 days unless flagged for safety investigations or legal reasons. This move nullifies prior zero-data-retention agreements with enterprise customers, effectively overriding existing bespoke privacy terms. The policy coincides with the June rollout of Claude Fable 5 and Claude Mythos 5, marking a broader shift in how enterprise data is managed for frontier AI capabilities. OpenAI has begun testing a competing approach known as private safety processing to avoid retaining customer data, signaling a wider industry trend toward privacy-preserving safety measures. Anthropic is also exploring a newer safety system that would let customers retain data on their own cloud infrastructure while still maintaining the 30-day retention window, a development that could ease enterprise concerns while the company continues to work with regulated industries.