Microsoft Blocks Powerful AI Over Data Fears

Microsoft has temporarily halted employee access to Anthropic's potent Claude Fable 5 AI model. This decision arises from concerns over Anthropic's mandatory data retention policy for its Mythos class models, including Fable 5. The policy requires storing user prompts and AI outputs for 30 days, and potentially up to two years if flagged for safety investigations. These requirements conflict with Microsoft's stringent enterprise data handling protocols.

This restriction addresses fears that employees might inadvertently share sensitive corporate or customer information with the AI. Fable 5, launched after the highly capable Mythos model, is considered safe for general use but its advanced cybersecurity and vulnerability discovery features are noteworthy.

Coinciding with Microsoft's cancellation of internal Claude Code licenses and a shift towards its own GitHub Copilot CLI tools, this action highlights a broader strategy. Microsoft appears to be prioritizing in-house solutions, though it remains open to leveraging third-party partners like Anthropic when advantageous.

Despite the temporary ban, internal demand for Fable 5 remains high. Its performance across various testing categories, including agentic coding and cybersecurity, reportedly surpasses leading models such as GPT-5.5, Gemini 3.1 Pro, and Opus 4.8, suggesting that internal pressure could lead Microsoft to reconsider its stance.