Microsoft blocks Claude Fable 5 integration over 30-day data retention policy

Microsoft has stopped employees from using Anthropic's Claude Fable 5 model, particularly blocking its integration into GitHub Copilot pending an internal security review. The root cause is Anthropic's newly updated policy for its Mythos-class models, which mandates that all user prompts and generated outputs be retained for 30 days for safety monitoring. Microsoft's security team flagged this as incompatible with internal data handling requirements, especially for code generation tools where proprietary code could be exposed.

This is not Microsoft's first restriction on Anthropic's tools. Earlier this year, the company canceled most of its internal licenses for Claude Code, another Anthropic assistant. The pattern reflects a broader tension: as enterprise AI adoption accelerates, companies like Microsoft are increasingly wary of third-party AI models that store or process sensitive data on external servers. The decision could also signal a push toward Microsoft's own AI models (e.g., Copilot powered by OpenAI) over rival offerings.