Microsoft limits Claude Fable 5: concerns over data

Microsoft has recently decided to limit the use of Claude Fable 5, the latest artificial intelligence model from Anthropic, for its employees. This decision stems from new data retention requirements imposed by Anthropic, which have raised concerns within the tech company.

Launched yesterday, Claude Fable 5 is the first model in Anthropic's Mythos class. Although it has been quickly integrated for clients of GitHub Copilot and Foundry, it remains absent from the model selector used by Microsoft employees for internal versions of GitHub Copilot. In contrast, other Claude models, which comply with Zero Data Retention (ZDR) rules, remain available internally.

Microsoft's legal teams are currently assessing the implications of Anthropic's new data retention requirements. The primary concerns revolve around the security of customer data and confidential information. It is still unclear whether Microsoft's legal teams will give the green light for the internal use of Claude Fable 5.

Claude Fable 5 indeed requires the retention of prompts and outputs for 30 days for the operation of its new security classifiers. In some cases, this data may be retained for up to two years if flagged as violating Anthropic's usage policy.

This AI model arrives just weeks after Anthropic stated that this family of models was so effective in cybersecurity tasks that it was too dangerous to be released publicly. These security measures have led to significant changes in data management, which explains Microsoft's hesitations.

We reached out to Microsoft for a comment regarding this situation, but the company did not respond in time for publication.