Microsoft Restricts Claude Fable 5 Internally as Anthropic Retains Prompts for 30 Days
Updated
Updated · The Verge · Jun 10
Microsoft Restricts Claude Fable 5 Internally as Anthropic Retains Prompts for 30 Days
3 articles · Updated · The Verge · Jun 10
Summary
Microsoft has kept Claude Fable 5 out of the internal GitHub Copilot model picker for employees while its legal teams review Anthropic’s revised retention terms.
30-day prompt-and-output retention is required for Fable 5’s new safety classifiers, and Anthropic can keep flagged interactions for up to two years, raising concerns over customer and confidential data.
That restriction applies to Microsoft’s internal use, not its external products: the company has already made Claude Fable 5 available to GitHub Copilot and Foundry customers.
Other Claude models remain available inside Microsoft because they still run under zero-data-retention rules, leaving Fable 5 as the exception.
Fable 5 is Anthropic’s first broadly released Mythos-class model, arriving weeks after the company said the family was too capable in cybersecurity tasks to release publicly without added safeguards.
Why is Microsoft selling an AI model that its own legal team deems too risky for internal use?
Is the price of next-gen AI safety the end of corporate data privacy as we know it?
Anthropic’s 30-Day Data Retention on Claude Fable 5 Triggers Microsoft Ban and Industry-Wide AI Security Debate
Overview
On June 10, 2026, Microsoft imposed an immediate internal ban on employee use of Anthropic's Claude Fable 5 due to serious data governance concerns. The core issue is Anthropic’s mandatory 30-day data retention policy for all prompts and outputs, which directly conflicts with Microsoft’s strict internal security standards and its commitment to protecting sensitive customer and proprietary information. This move highlights Microsoft’s broader worries about deploying powerful 'Mythos-class' AI models internally, especially as Anthropic has warned that advanced AI could soon self-improve without human oversight. The situation underscores the growing tension between rapid AI innovation and the need for robust data security.