- Mythos-based Claude Fable 5 is so powerful, Anthropic mandates data retention for safety monitoring
- Microsoft is worried about insiders sharing confidential information with Anthropic's model
- Workers told to use Microsoft alternatives, but third-party partner models still available
Microsoft has reportedly restricted employees from using Claude's recently launched Fable 5 model while legal and compliance teams review data handling.
Per Anthropic's latest retention policy for the Mythos class of models, which includes Fable 5, the company requires prompt and output retention for safety monitoring, which goes against enterprise-grade policies that customers like Microsoft have in place.
Anthropic stressed it needs to keep prompts and outputs for 30 days, but content flagged by its safety systems can be retained for up to two years if it's needed for investigation or enforcement.
Microsoft halts Claude Fable 5 use over data retention concerns
The current restriction, which applies to Microsoft workers, is in place over concerns that insiders may share sensitive information like customer information, corporate details and other confidential business information.
Fable 5 was launched following the huge success of Mythos, which was praised for its strong cybersecurity and vulnerability-discovery capabilities.
However that model was only launched to a select group of partners over concerns it could be misused by attackers, hence the subsequent launch of the tamed-down Fable variant which the company describes as "safe for general use."
Though it's likely unrelated, Microsoft's push to halt Fable 5 deployment follows recent news that the company was cancelling internal Claude Code licenses, moving its developers to its own GitHub Copilot CLI tools.
Together, the two shifts reflect Microsoft's desire to prioritize internal tools over third-party alternatives, likely due to cost benefits, but a clear willingness to use systems from Anthropic also suggests the company is willing to lean on partners when necessary.
"It’s not yet clear whether Microsoft’s legal teams will clear Claude Fable for internal use," The Verge added.
Though a temporary ban is in place, internal demand could force Microsoft to change its tune. Mythos 5 and Fable 5 outperform key frontier models like GPT-5.5, Gemini 3.1 Pro and Opus 4.8 across 13 testing categories, including agentic coding, knowledge work and cybersecurity.
