Due to data retention policies, Microsoft restricts employees from using Claude Fable 5 internally.

According to Dynamics Beating monitoring, due to concerns about data retention risks in Anthropic's new release model, there are reports that Microsoft has restricted employees from using Claude Fable 5 internally. While Microsoft has rolled out Claude Fable 5 to external GitHub Copilot and Foundry customers, the internal version used by employees does not provide Claude Fable 5.

The Microsoft legal team is evaluating Anthropic's data retention terms of service. In contrast to the Claude model internally provided by Microsoft, which follows the Zero Data Retention (ZDR) rule, Claude Fable 5 requires data retention to operate a brand-new secure classifier. Anthropic will retain users' input prompts and output for 30 days. If the content is deemed to violate usage policies, the related data could potentially be stored for up to 2 years, raising concerns at Microsoft about customer data and confidential information leakage.

As the first public version of an Anthropic Mythos-level model, Claude Fable 5 was previously assessed as too powerful for release in cybersecurity tasks. To mitigate security risks, Anthropic introduced a prompt-protected mechanism, which directly led to the data retention requirement.