Microsoft has restricted employee access to Anthropic’s newly released Claude Fable 5 AI model this week while its legal teams assess the implications of the startup’s revised data retention policy, raising fresh questions about how advanced AI systems handle sensitive corporate information.
The Verge reported on 10 June that Microsoft has removed Claude Fable 5 from the internal GitHub Copilot tools available to employees, despite making the model available to external customers through GitHub Copilot and Microsoft Foundry. According to the report, Microsoft’s concerns centre on the treatment of customer data and confidential information under Anthropic’s new retention requirements.
Anthropic launched Claude Fable 5 on 9 June as the first broadly available model in its Mythos family, describing it as its most capable publicly accessible AI system to date. Microsoft promoted the model’s availability on Azure and Foundry the same day, highlighting its ability to support long-running coding, research and enterprise workflows.
According to the report, Microsoft has told employees that its legal teams are evaluating the changes before deciding whether the model can be approved for internal use. Microsoft declined to comment to The Verge, while Reuters reported that neither Microsoft nor Anthropic immediately responded to requests for comment.
The dispute centres on Anthropic’s decision to require data retention for Mythos-class models to support safety monitoring systems. Under the policy, prompts and generated outputs are retained for 30 days across all platforms where the models are offered. Anthropic may keep data for up to two years if its trust and safety systems flag content as potentially violating company policies.
Anthropic said the retained data would be used only for safety-related purposes and not to train future AI models. In a statement accompanying the launch, the company said the information would help it defend against “complex and novel attacks” and improve the accuracy of its safety systems.
The development highlights growing tensions between increasingly powerful AI models and enterprise requirements for data protection. Zero data retention arrangements have become a key requirement for many corporate customers handling confidential information, and Claude Fable 5’s new safeguards appear to have created compliance challenges even for one of Anthropic’s closest commercial partners.
The issue emerges days after Microsoft AI chief Mustafa Suleyman publicly criticised Anthropic’s approach to AI consciousness. Speaking on The Verge’s Decoder podcast, Suleyman said it was “really, really dangerous” for developers to speculate about whether AI systems might be conscious, describing such discussions as a “philosophical failing”.
