Data Risks: Anthropic Models Enabled by Default in M365 Copilot, Starting Jan. 2026

Starting in Jan. 2026, Anthropic’s Claude models will be enabled in tenancies by default for use by some features of Microsoft 365 Copilot, potentially posing a risk for commercial customers with regulatory or data residency requirements.

Although Anthropic is a sub-processor for Microsoft and is now covered by Microsoft’s data protection safeguards, data is still sent to Anthropic’s datacenters for processing which are mostly in AWS and GCP data centers and mainly in the U.S. The use of Anthropic’s processing may not be acceptable for some regulatory or internal policies, such as data sovereignty requirements of global companies.

Customers can disable Anthropic’s models in the M365 Admin center in Copilot Settings. There is no loss of functionality since Claude models are optional to use in M365 Copilot features and don’t affect the default OpenAI model. However, the Anthropic Office agents for Word, Excel and PowerPoint will not be available and Microsoft doesn’t offer corresponding agents.