Microsoft Tells Employees To Avoid Anthropic's Claude Fable 5 Over Data Retention Concerns

Microsoft has restricted its employees from using Anthropic's newly launched Claude Fable 5, even as the company simultaneously offers the same model to its enterprise customers through GitHub Copilot and Microsoft Foundry. The restriction, first reported by The Verge, stems from concerns over Anthropic's updated data retention policy for its Mythos-class AI models.

What changed with Claude Fable 5?

Anthropic launched Claude Fable 5 on June 9, as a public version of its Mythos AI model, describing it as the most powerful model it has made available for wider use, with guardrails barring its use in high-risk areas such as cybersecurity.

Alongside the launch, Anthropic updated its data retention policy. Under the new policy, prompts submitted and outputs generated are retained for 30 days for trust and safety purposes on every platform where Mythos-class models are offered. Anthropic retains inputs and outputs for up to two years if they are flagged by its trust and safety classifiers as violating its usage policy.

Why Microsoft's lawyers pushed back

For Microsoft, which handles sensitive customer data and proprietary code, the retention window raises compliance and confidentiality concerns. The company has told employees that its lawyers are evaluating whether these requirements align with Microsoft's own standards for data protection.

The block applies specifically to the model picker that Microsoft employees use for internal versions of GitHub Copilot. All other Claude models remain available because they operate under Zero Data Retention rules.

Anthropic's position

Anthropic has said the new 30-day retention policy applies to both first- and third-party surfaces, adding that it will ensure data deletion after 30 days in almost all cases, and that it will not use this data to train new Claude models or for any purpose unrelated to safety.

The irony for Microsoft

It remains uncertain whether Microsoft's legal teams will clear Claude Fable 5 for internal use. The situation highlights a broader tension in enterprise AI adoption - a model can be commercially available and legally unapproved at the same time, even within the same company selling it.