copilotclaudegdprdpiaeu-data-boundary

Claude in Copilot: what the EU Data Boundary means for DPIAs

November 17, 20269 min readPixel Management

This article is also available in Dutch

On 5 November 2026 Microsoft announced that Anthropic's Claude models in Copilot will process data inside the EU Data Boundary for EU and EFTA customers "by the end of 2026". Until then, that processing happens outside Europe.

Claude in Copilot is today processing by a Microsoft subprocessor outside the EU Data Boundary, and by the end of 2026, according to Microsoft, it becomes processing inside that boundary under the same terms. For a DPIA, that changes one line in a fundamental way: where the model runs. Anthropic's role and the contract stay the same. What doesn't stay the same for every tenant is the default setting, and that's where the risk sits.

What Microsoft announced on 5 November 2026

Charles Lamanna, who runs Copilot at Microsoft, wrote in the Copilot blog announcement that organisations in the EU and EFTA will use Claude models "with processing within the EU Data Boundary, where their data is already stored". Three lines in that post matter to an admin:

  • "The setting is disabled by default in the EU and EFTA."
  • "Organizations already using Claude models in Copilot won't need to take any action."
  • The EU Data Boundary commitments "are included in our contractual terms, with no opt-in or additional purchase needed".

There's no firm date. Microsoft writes "Once available" and promises more detail as availability gets closer. A blog post also isn't a contract change. Microsoft's own EU Data Boundary overview says the services covered by the boundary are identified in the Product Terms, and what counts is the text in force when the processing happens.

Where does Anthropic sit in the chain today?

Anthropic is a Microsoft subprocessor for Claude in Microsoft 365. The Learn page Claude models from Anthropic in Microsoft Online Services, updated on 4 November 2026, says the Microsoft Product Terms and the Data Protection Addendum (DPA) apply to Claude used through Microsoft's enterprise online services, and Enterprise Data Protection covers it too.

The same page is just as direct about location. Claude models in Microsoft Copilot, Researcher, Copilot Studio, Power Platform and Copilot in the Microsoft 365 apps "are currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments". The Copilot privacy page repeats it in a separate note. Your Copilot conversations are stored in the EU. The model's compute isn't.

For a controller, that's a transfer outside the EEA. Article 44 of the GDPR only allows such a transfer when controller and processor meet the conditions of Chapter V. Which transfer mechanism Microsoft relies on for this subprocessing is in your own DPA, not in a blog. What EU data residency with a US provider does and doesn't settle legally is covered in EU data residency with US AI providers.

The exception: models with data retention

Not every Claude model in Copilot falls under Microsoft's contract. Models Microsoft labels "Anthropic models with Data Retention", such as Claude Fable 5, sit behind a separate switch. For those, the same Learn page says Anthropic acts as an "independent processor", Anthropic's own Commercial Terms and Data Protection Addendum apply, and Anthropic keeps most inputs and outputs for up to 30 days. If Anthropic's trust and safety classifiers flag something, content can be kept for up to two years and classification scores for up to seven. The 5 November announcement doesn't mention these models.

Claude in Copilot before and after the change

The table sets the position on 12 November 2026 against what Microsoft announced for the end of 2026.

ItemNow (12 November 2026)After the announced change
Where the model processesOutside the EU Data BoundaryInside the EU Data Boundary
Anthropic's roleMicrosoft subprocessorMicrosoft subprocessor
TermsMicrosoft Product Terms and DPAUnchanged, EUDB commitment contractual
Default in EU and EFTANo users, but see the exception below"Disabled by default"
Action for existing useOpt in again if you'd previously enabled it under Anthropic's own termsNo action needed
Models with data retentionAnthropic as independent processor, own terms, 30-day retentionNot mentioned in the announcement
In-country processingExcludedNot mentioned

Is Claude really off in your tenant?

Claude in Copilot isn't off in every EU tenant. The Learn page Enable Claude models only for Copilot experiences in Microsoft 365 apps, updated on 9 November 2026, says the All users for Microsoft 365 apps only option is enabled by default for tenants in the EU, EFTA and UK created after 25 March 2026. Claude then runs in Copilot in Word, Excel and PowerPoint, and per that same page "data processing for these models occurs outside of the EU Data Boundary".

That's the trap. "Off by default in the EU" holds for the general setting, which sits at No users. It doesn't hold for every tenant created after 25 March. If you set up your tenant this year, or a new tenant for a subsidiary, your staff may already be sending documents to Claude for processing outside the EU without anyone having chosen that. For older tenants, Microsoft recommends checking the setting yourself.

You'll find it in the Microsoft 365 admin center under Copilot, Settings, View all, AI providers operating as Microsoft subprocessors. Under Available subprocessors for your organization you pick Anthropic, and then decide who gets access: all users, the Microsoft 365 apps only, or specific users and groups. You need the AI Administrator or Global Administrator role. Until mid-September 2026 the apps option was called "Allow Anthropic models in Copilot experiences within Microsoft 365 apps", which helps if you're checking an older configuration document against the screen.

What changes in your DPIA and processing agreement?

Your processing agreement with Microsoft doesn't change. Anthropic was already a subprocessor under the DPA, and Article 28(2) GDPR requires the processor to inform you of any intended addition or replacement of other processors so you can object. For Anthropic, that moment has passed. Microsoft points to the Service Trust Portal for the subprocessor list, so record which version you reviewed.

Your DPIA does change. For a 50 to 500 person organisation, four passages are affected.

  1. Description of the processing. Record which Copilot parts use Claude: Copilot Chat, Researcher, the Office apps, Copilot Studio. Each can have its own setting.
  2. Location and transfer. Until the change is live: processing outside the EU Data Boundary, with the transfer mechanism from your DPA. After it: inside the boundary. Add a review date, say the first working day of January 2027, and the test: the line "currently excluded from the EU Data Boundary" is gone from the Learn page.
  3. Models with data retention. Record that they're off, or assess them as a separate processing with Anthropic as independent processor and retention from 30 days up to two years.
  4. Access. Record which security group may use Claude and why. Microsoft supports up to 999 assignments per AI provider, and access is set per provider, not per model.

If you're writing the Copilot DPIA from scratch, follow the steps in a DPIA for AI projects. For the contract itself, subprocessor terms belong with the contract terms for buying AI software.

Time saved

Save 4 hours per week on aligning DPIA passages with tenant settings for AI subprocessors

Set it now in the admin center

Don't wait for the end of 2026 to make a deliberate choice. Here's what we'd set for a client of this size today.

  1. Open AI providers operating as Microsoft subprocessors and record the current state. If All users for Microsoft 365 apps only is on without a decision behind it, switch it back or document the decision now.
  2. Limit Anthropic to a security group. A pilot group from one department without sensitive files is enough to learn what Claude adds.
  3. Keep Anthropic models with Data Retention off. That switch is off by default everywhere, and it should stay off until you've reviewed Anthropic's terms.
  4. Check Copilot Studio and Power Platform separately. Once Claude is enabled in the Microsoft 365 admin center, Microsoft says further controls appear in the Power Platform admin center.
  5. If you'd enabled Claude earlier under Anthropic's own terms, enable it again through the subprocessor route. Microsoft says those organisations need to opt in again.

If you've deliberately enabled Claude on the subprocessor route, Microsoft says you won't need to do anything when the change lands. The processing moves on its own.

When to keep Claude in Copilot switched off

Keep Claude in Copilot off for groups handling special category data for as long as processing happens outside the EU Data Boundary. Think HR with sickness records, a healthcare unit, or a team handling criminal records data. Article 9 GDPR already sets a higher bar there, and an extra transfer makes the DPIA balance harder, while those groups can keep using Copilot with the other models.

If you work under professional secrecy, as lawyers, notaries, accountants and doctors in the Netherlands do, location is only one question. Professional rules also ask who can see the data and for how long it's kept, and the data retention models keep it at Anthropic for up to 30 days. What the Dutch professional bodies say about that is in AI and professional secrecy.

There's a business reason to wait, too. Claude in Copilot only pays off if it does a task better than the default model, and you'll only know that after a test on one department's real documents. What a copilot delivers per department is in using AI copilots in business. If your most sensitive knowledge is the core of the business, a private AI environment for that part is easier to defend than any data boundary. The wider rules for AI use in the Netherlands are in our overview of AI legislation in the Netherlands and the EU.

Service

Learn more about AI consulting?

View service →

Curious how much time you could save?

Request a free efficiency audit. We'll analyze your processes and show you where the gains are, no strings attached.