Microsoft restricts Claude Fable employee use over data retention fears

Microsoft has blocked internal employee access to Anthropic's just-released Claude Fable 5, the first Mythos-class AI model, over significant data retention concerns. According to sources, the model has been removed from the model picker in Microsoft's internal version of GitHub Copilot, though it remains available to external customers via GitHub Copilot and Foundry. All other Claude models continue to operate under Microsoft's Zero Data Retention (ZDR) rules, which the company has long required to protect customer data and confidential information. Microsoft's legal teams are currently evaluating the new data retention requirements imposed by Anthropic, and it remains unclear whether Claude Fable will be cleared for internal use.

Claude Fable 5's data retention policy is a direct consequence of Anthropic's advanced safety classifiers, which require storing user prompts and outputs for 30 days to monitor for misuse. In cases where content is flagged as violating usage policy, data can be retained for up to two years. This creates a fundamental conflict with Microsoft's enterprise security standards, which typically mandate that no user data be stored by third-party AI providers. The restriction highlights the growing tension between powerful AI capabilities that demand operational data and the strict privacy requirements of large enterprise customers. Anthropic previously warned that the Mythos class of models was too dangerous for public release, leading to these safety-focused but controversial data retention changes.