Microsoft curbs use of Claude Fable 5 over data retention concerns

Microsoft has curtailed employee access to Anthropic’s new AI model, Claude Fable 5. As reported by Reuters and The Verge, the company is reviewing whether the model’s data retention terms meet its internal standards for privacy, security, and protection of confidential information. That’s according to Reuters, among others.

The move comes shortly after the debut of Claude Fable 5, the most powerful publicly available model in Anthropic’s new Mythos lineup. Marketed as a major leap for software development, analysis, and complex AI tasks, the model’s data storage rules are nevertheless giving large enterprises pause.

At the heart of the issue is how Anthropic handles user data.

By default, Anthropic stores prompts and responses from Mythos models for 30 days for trust-and-safety purposes. If potential policy violations are detected, that data can be retained for up to two years for further investigation.

For Microsoft, that’s a potential risk. Employees at major tech firms routinely handle customer data, internal documents, and other sensitive information. Even if stored data isn’t used for training, prolonged retention raises legal, compliance, and privacy concerns.

According to The Verge, Microsoft’s legal teams have temporarily restricted internal use of Claude Fable 5 while they assess the policy’s implications. It’s not yet clear whether the model will ultimately be cleared for internal deployment.

What makes this stand out is that earlier Claude models could run under Zero Data Retention terms—where prompts and outputs aren’t stored after a request is processed.

Claude Fable 5 is the exception. For its Mythos-class models, Anthropic has introduced mandatory retention to enable new safety systems and abuse detection.

It’s a sign that the latest, most capable AI systems increasingly collide with established enterprise policies on data governance and compliance.

Microsoft’s decision doesn’t signal a break with Anthropic or other third-party AI providers. In fact, Claude Fable 5 remains available to customers through various Microsoft platforms, including developer tools and AI services. For now, the restriction primarily targets internal employee use.

Still, the episode highlights a larger shift in the AI market. As models get more powerful, the conversation is moving from raw performance to governance, data security, and legal risk. For large companies, it’s not just how good a model is that matters—it’s what happens to the data you put in, who can access it, and how long it’s kept.

Those questions are increasingly decisive for enterprise adoption of advanced AI.