Microsoft is limiting employee access to Anthropic’s Claude Fable 5 while its legal teams review new data retention requirements tied to the model, according to a report from The Verge.

The restriction comes just a day after Anthropic unveiled Claude Fable, its first model in the Mythos family, and as Microsoft continues to offer the system to external customers through GitHub Copilot and its Foundry platform. Internally, however, the model is not appearing in the picker Microsoft employees use for its own GitHub Copilot tools.

Other Claude models remain available to Microsoft staff because they operate under zero data retention rules, which do not require the provider to keep prompts and outputs. Claude Fable 5 is different. Anthropic’s new safety setup depends on retaining data, meaning prompts and responses are stored for 30 days. In some cases, if content is flagged as a policy violation, it can be held for as long as two years.

That retention policy appears to be the core issue for Microsoft. The company is concerned about how customer information and confidential material might be handled if employees use the model internally. According to the report, Microsoft has told employees that its legal teams are examining Anthropic’s updated requirements, but it is still unclear whether the company will approve Claude Fable 5 for broader internal use.

Microsoft declined to comment.

The move highlights the tension between rapid deployment of new AI systems and the data handling rules that govern them. Anthropic has positioned Claude Fable 5 as a more advanced model with strong safety controls, but those controls come with stricter retention terms than the zero-retention framework Microsoft has used with other Claude products.

Those concerns are especially sensitive for Microsoft, which handles large volumes of proprietary code, business data, and customer-related information through its internal AI tools. Even if the model is already available to paying customers, companies often apply different standards to employee use, particularly when legal or compliance questions are involved.

Anthropic’s Mythos models have also drawn attention because of their capabilities and the caution around their release. The company recently said the family was so capable in cybersecurity-related work that it was considered too risky for public release at one point. For Claude Fable 5, Anthropic has added safeguards intended to reduce misuse, and the resulting system includes the retention policy now under review at Microsoft.

For now, the situation leaves Microsoft in a familiar position for major AI customers: balancing access to the latest models against privacy, compliance, and legal considerations. The company has moved quickly to support Claude Fable 5 for cloud customers, but internal use may depend on whether its legal review concludes that Anthropic’s retention approach is acceptable.

As of now, Microsoft employees cannot broadly use Claude Fable 5 in internal Copilot tools, and the company has not said when that might change.