Amazon Web Services has tightened its Bedrock policy for Anthropic’s Mythos‑class models, including Fable 5 and Mythos 5. The change forces a 30‑day data retention period on all traffic that passes through these models. By moving data beyond AWS’s security boundary, Anthropic can monitor for abuse across multiple interactions for developers using conversational AI at scale to improve compliance standards.

The policy applies to any request that reaches the Mythos‑class endpoint, with data automatically purged after the retention window unless retained for a safety probe or legal hold. Anthropic’s support page notes that deletion occurs within 30 days, except in rare cases where investigations or regulations demand longer storage. This mechanism signals a shift toward proactive misuse detection in LLMs.

Developers choosing Bedrock for Anthropic models must enable this setting, accepting that user inputs will leave AWS’s protected ecosystem. The retention model balances privacy with safety, allowing pattern analysis that single prompts cannot reveal. Organizations deploying conversational AI at scale will need to adjust their data governance to accommodate the 30‑day window and the potential for cross‑session monitoring across applications.