Microsoft blocks Claude Fable 5 for its own employees over data policy

Anthropic yesterday released Claude Fable, its first Mythos-class model, but problems are already surfacing inside Microsoft. The company has restricted employee access to Claude Fable 5 because of Anthropic's new requirements for data storage and processing.

A fast launch for customers

Microsoft quickly rolled out Claude Fable 5 to its customers on GitHub Copilot and Azure Foundry, with updates shipping the same day. The public release went smoothly, and the model became available in the model picker for all users. This suggests Microsoft and Anthropic coordinated well before the launch.

For GitHub Copilot and Foundry customers, access to Fable is a significant upgrade. The model promises better performance than previous Claude versions, and Microsoft's enthusiasm was justified.

But inside the company, a ban

Inside Microsoft itself, the situation is different. According to The Verge's sources, Claude Fable 5 is missing from the model picker in the internal version of GitHub Copilot used by employees. Microsoft staff cannot select Fable, even though the company has just offered it to its customers.

The decision was made by Microsoft itself, not by Anthropic. All other Claude models, including older versions, remain available for internal use. The company chose not to give its own team access to the newest model.

Zero Data Retention: the main point of disagreement

The key reason is data handling policy. All Claude models except Fable run under a Zero Data Retention (ZDR) regime. This means Anthropic does not store or use data passed through the API for model training or other purposes.

The ZDR commitment is a critical clause in Microsoft's contract with Anthropic. The company requires this regime for all enterprise customers, including Azure Foundry. Claude Fable ships with different data handling terms that are not compatible with the ZDR policy. Microsoft rejected them.

"Employee data requires maximum protection"

Why this matters

For Microsoft, this is not just a technical question but one of risk management:

• Employee data (source code, documents, technical communications) is more sensitive than customer data
• Storing corporate information requires the highest level of protection
• ZDR was chosen as the only acceptable regime for internal use
• Departing from ZDR would set a precedent for other models and vendors

Microsoft may revisit the terms, but for now the block remains.

What this means

The conflict shows the growing tension between the desire to use the newest AI models and the need to protect sensitive corporate data. Microsoft may be the first large company to face this dilemma with Fable, but it will certainly not be the last.

The question remains open: will Anthropic revise its terms, or will Microsoft accept a new level of risk?