The review highlights growing scrutiny of AI governance as companies balance advanced capabilities with data security and compliance demands.
Microsoft has restricted employees from using Anthropic’s newly launched Claude Fable 5 AI model internally, citing concerns over the startup’s data retention practices, according to a report by The Verge.
The move comes a day after Anthropic introduced Claude Fable 5, the public-facing version of its Mythos-class model.
While Microsoft has already made the model available to customers through GitHub Copilot and Microsoft Foundry, employees inside the company are currently unable to access it as legal and compliance teams review the implications of Anthropic’s policies.
The restriction reportedly stems from a key difference between Claude Fable 5 and Anthropic’s earlier models. Most Claude models available to Microsoft employees comply with Zero Data Retention (ZDR) requirements, meaning user prompts and outputs are not stored.
Claude Fable 5, however, retains user data for 30 days to support Anthropic’s new safety monitoring systems. Content flagged by those systems can be stored for up to 2 years, raising concerns about the handling of sensitive customer and corporate information.
Microsoft’s legal teams are evaluating whether the model can meet the company’s internal standards for protecting confidential data before broader employee access is approved, according to the report.
Claude Fable 5 is Anthropic’s most capable publicly available model to date. The company has described it as particularly strong in software engineering, research, and complex knowledge work, while also incorporating safeguards for sensitive areas such as cybersecurity, biology, and chemistry.
Also Read: Is AI Becoming America’s Next Security Crisis?
Anthropic introduced those safeguards because the underlying Mythos model demonstrated advanced capabilities that the company previously considered too risky for broad release. The public-facing Fable 5 version includes restrictions that route certain high-risk requests to less capable models, while Claude Mythos 5 remains available only to a limited group of vetted organizations through Anthropic’s Project Glasswing program.
The development also follows Microsoft’s recent efforts to reduce its internal reliance on certain Anthropic tools. Last month, the company began canceling many internal Claude Code licenses and directing engineers toward GitHub Copilot CLI, Microsoft’s own AI-powered coding assistant.
For now, Microsoft employees can continue using other Claude models that comply with the company’s data-handling requirements, while Claude Fable 5 remains under review.
