Anthropic requires 30 day data retention for Fable and Mythos (support.claude.com)

Anthropic has announced a new policy requiring a 30-day data retention for prompts and outputs generated by its Mythos-class models, effective June 9, 2026. This decision reflects Anthropic's commitment to safety and trust, as the company aims to responsibly deploy models with advanced capabilities that can be misused. The policy mainly affects organizations using Claude’s capabilities with zero data retention (ZDR), while consumer plans remain unaffected under existing terms. This proactive approach will help detect patterns of misuse, such as sophisticated prompt attacks that become visible only through aggregated data analysis. The retention policy is designed to augment safety measures for advanced models, particularly addressing potential threats in cyber and bio domains. Anthropic emphasizes stringent data protection protocols, stating that any employee access to this retained data will be highly regulated and logged. The retention strategy also allows organizations to implement customer-managed encryption keys for added security. By maintaining transparency and privacy controls, Anthropic reinforces its security framework while enhancing its ability to identify and mitigate potential misuse of its powerful AI models.