Anthropic plans new enterprise data controls for Claude customers

Anthropic just told its enterprise customers something they really didn’t want to hear: every prompt and every output from its most powerful new AI models will be stored for 30 days. No exceptions, no opt-outs, no matter what your previous contract said. The policy, announced alongside the launch of Claude Fable 5 and Claude Mythos 5 on June 9, applies universally across first-party and third-party platforms. For a company that built its brand on being the responsible AI lab, this is a fascinating gamble: prioritizing safety monitoring over the zero-retention guarantees that made it attractive to privacy-conscious enterprises in the first place. What the policy actually does Under the new terms, Anthropic will log all prompts sent to and outputs generated by its Mythos-class models, which include both Fable 5 and Mythos 5. The retained data is earmarked exclusively for safety monitoring, specifically detecting novel attacks and potential vulnerabilities in these frontier systems. Anthropic has been explicit that the logged data will not be used for model training. After 30 days, it gets automatically deleted, unless a specific interaction has been flagged for a safety investigation...