Microsoft Puts Brakes on Anthropic's Latest AI Model Over Data Handling Rules

Anthropic's debut of Claude Fable 5 - the first model in its Mythos family - has hit a snag inside Microsoft. According to sources, the tech giant is restricting employee access to the model due to new data retention policies that conflict with Microsoft's internal safeguards.

While Microsoft quickly pushed Fable 5 out to customers through GitHub Copilot and Foundry, the model is conspicuously absent from the internal model picker used by employees. Other Claude models remain available under Microsoft's standard Zero Data Retention (ZDR) rules, but Fable 5 operates differently.

Anthropic requires that prompts and outputs be stored for 30 days to power its safety classifiers. In cases where content violates usage policies, data can be held for up to two years. That's raising red flags at Microsoft, where legal teams are now evaluating whether these terms pose a risk to customer data and confidential information.

Claude Fable 5 arrives just weeks after Anthropic described its Mythos class as so capable in cybersecurity that it was too dangerous to release publicly. The company added safeguards to make Fable 5 safer, but those protections come with these data retention strings attached.

Microsoft did not respond to a request for comment by press time. For now, employees will have to wait while the lawyers sort out whether the model gets the green light for internal use.

Source: The Verge