Microsoft has reportedly blocked employee access to Anthropic’s newly launched AI model, Claude Fable 5. The tech giant cited concerns over the model’s data-retention practices, even as it continues offering the same technology to enterprise customers through its platforms.
According to reports, Microsoft employees can no longer access Claude Fable 5 through the company’s internal version of GitHub Copilot. Other Claude models remain available for internal use, as they operate under Anthropic’s Zero Data Retention (ZDR) policy.
The restriction stems from changes introduced with Claude Fable 5, Anthropic’s first model in its new Mythos family. Unlike previous Claude models, Fable 5 requires prompts and outputs to be temporarily stored to support additional safety mechanisms. Under Anthropic’s policies, user interactions are retained for 30 days, while content identified as potential policy violations may be preserved for up to two years.
The development highlights the growing tension organisations face between adopting advanced AI capabilities and ensuring compliance with internal data governance standards. For a company such as Microsoft, which manages sensitive customer information and proprietary business data, data handling requirements remain a critical consideration.
The move is particularly noteworthy given Microsoft’s expanding relationship with Anthropic. The two companies have collaborated to bring Claude models to Microsoft’s AI ecosystem, including offerings through Azure AI Foundry and GitHub Copilot. Customers using these services continue to have access to Claude Fable 5 because Microsoft applies its own retention policies within those environments.
Internally, however, Microsoft’s legal and compliance teams are reportedly still assessing whether the model meets the company’s standards for employee use. Until that review is completed, staff will remain limited to Claude models that qualify under Zero Data Retention provisions.
The episode underscores a broader challenge emerging in enterprise AI adoption: organisations are increasingly evaluating not only a model’s performance, but also the underlying policies governing data storage, security and regulatory compliance.
Anthropic had previously indicated that its Mythos family possessed advanced cybersecurity capabilities requiring enhanced safeguards before wider deployment. The additional retention measures introduced with Claude Fable 5 form part of those safety controls, but they have also become a potential barrier for businesses with stringent privacy requirements.
