Report: Microsoft Restricts Employees' Claude Access Over Data Retention

Microsoft's public-facing AI tools, like GitHub Copilot, now offer access to Claude Fable 5, but its internal tools restrict it, The Verge reports . Despite the new model being touted as excellent for coding, cybersecurity, and other high-level tasks, Microsoft's concerns over Anthropic's data-retention policies mean it's having to run it all by the lawyers first.

Claude Fable 5 is Anthropic's new AI model. It's built to be "Mythos Class," in that it's even more powerful than the usual Claude models. That seems to mean it's really good at coding but could, in theory, be used to create malware. To stop that, Anthropic has, it claims, put strong guardrails in place, after previously suggesting Mythos was far too powerful to release to the public .

As part of the Claude Fable 5 release, Anthropic has changed its data-retention policy to retain prompts and outputs for 30 days. But it can hold on to that data for up to two years if it's flagged for violating Anthropic policies. This could make Microsoft's sensitive data vulnerable, and it's investigating whether that's a risk it's willing to take before granting employees access.

Other Claude models operate without data-retention rules, making them easier for Microsoft to approve for internal use. Its GitHub Copilot and Foundry customers have access to Claude Fable 5, The Verge reports.

This is where the needs of AI companies to retain some responsibility over model output (and user input) meet corporate needs for privacy and security. As Microsoft lawyers hash out what's viable, it may be that some AI models simply aren't viable for major organizations that want to retain control over their data.

To make sure your chats with Claude are as secure as possible, check out these privacy tips .