Back to the blog

Claude in Microsoft 365 Copilot moves into the EU Data Boundary: the setting to check now

· By Bee Flow · 3 min read

Microsoft will process Claude models in Copilot inside the EU Data Boundary by the end of 2026. Admin blogs report that the models are then switched on for tenants that never made a choice. What is confirmed, what is not, and the setting to look at this month.

A control panel with three switches inside a dotted boundary line

On 5 November 2026 Microsoft announced that Claude models in Copilot will be processed inside the EU Data Boundary. In the announcement the company writes that by the end of 2026 organisations in the EU and EFTA can use Claude models across Copilot, including Copilot Cowork, with processing within that boundary. It gives no exact date and says more details follow closer to availability.

If your office works with Microsoft 365, this touches one admin setting. It is worth five minutes this month, because of a date that circulates among administrators: 9 December.

What Microsoft has confirmed

Copilot does not run on one model. Next to the OpenAI models, Microsoft offers Claude models from Anthropic. For those, Anthropic works as a subprocessor of Microsoft. Microsoft's documentation, Claude models from Anthropic in Microsoft Online Services, says the Microsoft Product Terms and the Data Protection Addendum apply to that use, unless a model carries the label "Anthropic models with Data Retention".

The same page, last updated on 10 November, describes the situation today. Claude models in Copilot, Researcher, Copilot Studio and Copilot in the Microsoft 365 apps are currently excluded from the EU Data Boundary. For customers in the EU, EFTA and the UK they are disabled by default, and the setting stands at "No users". Elsewhere in the commercial cloud Microsoft switches them on by default.

According to the announcement, tenants that already use Claude models do not have to do anything. Their processing moves into the boundary once that is available. A tenant that has not enabled them can do so at any time.

What is reported but not in Microsoft's public pages

Several administrator blogs describe a further step. One of them, Topedia, writes on 14 November that Microsoft will enable the Anthropic models on 9 December 2026 for EU and EFTA tenants that still have the default configuration. In that account a tenant where an administrator has explicitly disabled the setting keeps its choice.

We could not find that date in the announcement or in the documentation. Notices of this kind usually reach administrators through the Message Center in the Microsoft 365 admin center, which is not public. So we treat 9 December as likely and unconfirmed. Your own Message Center is the place to check it.

What to do this month

  1. Ask whoever manages your Microsoft 365 to open the admin center and go to Copilot, then Settings, then "AI providers operating as Microsoft subprocessors". Microsoft's documentation names the AI Administrator and Global Administrator roles for this.
  2. See what it says for Anthropic. Never touched means you are on the default.
  3. Decide, and record the decision. If you want Claude models off until you have looked into it, switch the setting off explicitly. If you want them, choose which users or groups get access.
  4. Search the Message Center for "Anthropic" and read what Microsoft sent your tenant. That text is the one that counts for you.
  5. Update your records. A new subprocessor behind a tool that handles client data belongs in your register of processing activities, and your AI policy should say whether colleagues may use these models.

One thing the documentation is clear about: the models labelled "Anthropic models with Data Retention" stand apart. They are off by default everywhere and need a separate opt-in. For those, Anthropic acts as an independent processor under its own terms.

Whether Claude in Copilot fits your firm is your own assessment. Processing inside the EU answers the question where the data is handled. It does not answer which company handles it, or which terms apply. We wrote about those questions in Is Microsoft 365 Copilot GDPR compliant? and in a data processing agreement for AI chat tools. This post is no legal advice.

Where Bee Flow fits

In Bee Flow the choice of model is a setting you make yourself. An administrator decides which providers the organisation uses: Claude, OpenAI, Mistral and other models hosted in the EU, or a model that runs locally on your own server. Bee Flow itself runs on your own server or on Bee Flow Cloud in the EU, and Privacy Shield blocks personal data before a message reaches the model, in every licence.

Bee Flow also works next to Microsoft 365, so this is no either-or. Bee Flow and Microsoft Copilot puts the two side by side, privacy and security shows where data goes, and you can plan a call to go through your own set-up.

#copilot #compliance #ai-at-work