Microsoft limits Claude Fable 5 use over Anthropic data retention rules

Microsoft moved quickly to offer Anthropic’s Claude Fable 5 to customers through GitHub Copilot and Microsoft Foundry, but it has also restricted internal employee use because the model requires Anthropic to retain prompts and outputs for 30 days to support safety monitoring. That split captures the tension now shaping enterprise AI: a model can be attractive enough for customer rollout and still fail an internal compliance test.

Anthropic launched Claude Fable 5 on June 9, describing it as the first model in its Mythos class and pricing it at $10 per million input tokens and $50 per million output tokens. The company says Fable requires 30-day data retention for safety monitoring, and GitHub’s model-hosting documentation says Anthropic retains prompts and outputs when Claude Fable 5 is used so safety classifiers can detect harmful use.

Microsoft’s own support materials draw a sharp line around those models. In Microsoft Online Services, Anthropic models with data retention must be separately enabled under Anthropic’s separate commercial terms and data processing agreement, while Microsoft says its Product Terms and Data Protection Addendum apply to other Anthropic models used through its enterprise services. Microsoft’s Foundry documentation adds that, for Anthropic Claude models there, Anthropic is the data processor for prompts and outputs, while Microsoft handles the deployment infrastructure and billing information.

That division explains why Microsoft can widen access for paying customers while drawing a harder internal boundary for employees. The practical lesson reaches beyond one model launch: procurement teams, compliance officers and legal departments are now deciding which frontier models can be used, where the data lives and who bears responsibility when retention rules change. In enterprise AI, trust is becoming a feature, and data governance is increasingly the gatekeeper.