I understand that Microsoft has been telling employees that its legal teams are evaluating changes to Anthropic’s data retention requirements. I’m told that the main concerns are around customer data and confidential information, and that it’s not yet clear whether Microsoft’s legal teams will clear Claude Fable for internal use.

Claude Fable 5 requires data retention to operate Anthropic’s new safety classifiers, meaning Anthropic retains prompts and outputs and deletes this data after 30 days. Some prompts and outputs can even be stored for up to two years if they’re flagged as violating Anthropic’s usage policy, so there are legal concerns around how Microsoft should be using Claude Fable 5.

Claude Fable 5 is the first broad release from Anthropic’s Mythos class of AI models, and it arrives just weeks after the company said the family was so capable at cybersecurity tasks that it was too dangerous to release publicly. Anthropic has put prompt safeguards in place to make Fable 5 less dangerous, leading to these challenging data retention changes.