Microsoft Blocks Employees From Using Anthropic's Claude Fable 5 Over Data Retention Concerns

Microsoft Pulls Fable 5 From Internal Tools

Microsoft has blocked its employees from accessing Anthropic's newly released Claude Fable 5 model through its internal tools, citing concerns over the AI startup's mandatory data retention policies, according to a report from The Verge. The restriction comes just one day after Anthropic launched Claude Fable 5, its most powerful publicly available model to date and the first in its Mythos class to be offered for general use.

What Triggered the Restriction

At the heart of the dispute is Anthropic's data retention policy for Mythos-class models, which took effect on June 9. The policy requires that all prompts and outputs be stored for 30 days for trust and safety purposes across every platform where the models are offered. Content flagged by Anthropic's safety classifiers as violating its usage policy can be retained for up to two years. Critically, the policy overrides the existing zero data retention agreements that enterprise customers, including Microsoft, had previously negotiated with Anthropic.

Microsoft has told employees that its legal teams are evaluating whether the new retention requirements are compatible with its internal data handling obligations, particularly around customer data and confidential information. As it stands, Claude Fable 5 does not appear in the model picker available to Microsoft employees through internal GitHub Copilot, though all other Claude models remain accessible under zero data retention rules.

Available to Customers, Not Staff

The situation has created an unusual split. Despite restricting its own staff, Microsoft has made Claude Fable 5 available to external customers. GitHub announced that the model is generally available for Copilot Pro+, Max, Business, and Enterprise users, though administrators must explicitly enable it since the policy is off by default. The model is also listed in Azure's AI catalogue.

Anthropic has pushed back on the implication that retention equals risk. According to its official documentation, retained data is not used to train new models and is deleted automatically after 30 days, except in cases involving active safety investigations or legal requirements. A small set of approved human reviewers can access flagged conversations through scoped tooling that prevents export or downloading.

Broader Context

The restriction arrives amid a broader recalibration of Microsoft's relationship with Anthropic's tools. Microsoft has separately been phasing out internal Claude Code licences by June 30 as it consolidates around its own Copilot CLI. Anthropic's models will remain available to Microsoft employees through Copilot CLI going forward, but the Fable 5 data retention question remains unresolved pending the legal review. The episode highlights the growing tension between AI safety obligations and enterprise confidentiality expectations as frontier models become more powerful.