Claude in UK Copilot Tenants From 9 December 2026: What Admins Must Decide

UK administrators of Microsoft 365 Copilot should record a decision on Anthropic's Claude models before 9 December 2026. The Register reported on 10 November that from that date, UK tenants whose administrators have not chosen a setting will have Claude access enabled, with the relevant data processed in the United States.

Two more changes arrived in the same fortnight. Microsoft said on 5 November that Claude processing for EU and EFTA tenants will move inside the EU Data Boundary by the end of 2026, and on 2 November it began applying unpublished usage limits to Copilot Chat for people without a Copilot licence. The Claude changes depend on the tenant's region, and the Chat limits depend on each user's licence. Microsoft Ignite runs from 17 to 20 November, so settle these decisions first.

What changed, and for which tenants

Microsoft's documentation says Claude models are already on by default "for most customers in commercial cloud (excluding EU/EFTA and UK)", with Anthropic acting as a Microsoft subprocessor. A subprocessor is a third party that Microsoft engages to process customer data under Microsoft's own contract with you.

The UK date rests on The Register's report. Microsoft's Learn page, as it stood on 10 November 2026, still listed the UK among the regions where the default is "No users", and the research for this article did not locate a Message Center notice for the change. UK admins should confirm the date in their own tenant's Message Center.

What "on by default" means in a UK tenant

The reported change applies when an administrator has not already chosen a setting. In that case, from 9 December users gain access to Claude in Copilot, and The Register reports, citing Microsoft's documentation, that Anthropic processes the relevant data in the United States. The wording implies that an existing choice is kept, although Microsoft has not publicly documented how it decides that a choice was made.

Enabling Claude makes it available to permitted users, and how it is then used varies by feature. Microsoft's documentation lists Microsoft Copilot, Researcher, Copilot Studio, Power Platform and Copilot in Microsoft 365 apps. In Word, Excel and PowerPoint the same page describes a setting that makes Anthropic the default model, so users there may not make a visible choice.

Where the setting lives

Microsoft documents the path as the Microsoft 365 admin center, then Copilot, Settings, View all, and "AI providers operating as Microsoft subprocessors". Anthropic sits under "Available subprocessors for your organization", and the task needs the AI Administrator or Global Administrator role. Access can go to all users or to named users and Microsoft Entra security groups, with up to 999 combined assignments per provider. Copilot Studio and Power Platform have further controls in the Power Platform admin center, so this one setting is not the whole configuration.

Newer tenants have an exception. EU, EFTA and UK tenants also have an option named "All users for Microsoft 365 apps only", which limits Claude to Word, Excel and PowerPoint, and Microsoft says it is enabled by default for tenants created after 25 March 2026. Such a tenant may already be using Claude in those apps, with processing that the same page says happens outside the EU Data Boundary. The Register also reports that Microsoft is retiring this apps-only option, and gives no date.

The better course for most UK organisations is to make the choice explicit before 9 December: off, or on for a named security group, with the reason written down. Check dependencies before choosing off, because Microsoft warns that some features are only available when Claude models are enabled. Letting the default apply is defensible where the organisation already accepts US processing by Microsoft's subprocessors for the same data and no customer contract restricts it. The case for wide access would be much stronger if Microsoft announced Claude processing inside the UK, which it has not done.

What the EU Data Boundary change covers, and what it leaves out

The EU Data Boundary is Microsoft's commitment to store and process EU and EFTA customer data inside that region, subject to limited exceptions that Microsoft documents. Claude has sat outside it. The Learn page still says Claude models "are currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments". The 5 November announcement says that by the end of 2026 EU and EFTA organisations can use Claude across Copilot with processing inside the boundary, and that tenants already using Claude will have processing "move into the EU Data Boundary automatically".

Microsoft's post says "The setting is disabled by default in the EU and EFTA", so nothing switches on by itself, apart from the apps-only default for newer tenants. It gives no exact date, and it does not say whether the stricter in-country processing commitments will cover Claude. The UK is outside the boundary, so this change does nothing for UK tenants. Until Microsoft confirms availability, an EU tenant that enables Claude is still using processing outside the boundary.

Processing location is one part of the assessment. Microsoft states that Claude in Copilot is covered by its enterprise data protection and that customer data is not used to train foundation models. Those are the vendor's commitments, and each organisation still has to judge whether they are enough. An EU organisation that kept Claude off for residency reasons should wait for confirmation in its own Message Center. A team that already opted in after a transfer assessment has no action to take, because Microsoft says the move is automatic.

The paperwork that may need an update

Microsoft says its Product Terms and Data Protection Addendum apply to Claude models under the standard arrangement, so there is no new contract to sign. A vendor register or record of processing that lists Copilot's subprocessors may need Anthropic and its processing location added. A data protection impact assessment written when Copilot used one provider may need an addendum covering the second provider and who has access. In countries with works councils, an agreement that describes how Copilot handles employee data may need a notice or an amendment, depending on its wording. These are questions for the data protection officer and employment counsel, and Microsoft's documents do not confirm any of them as mandatory.

Models that Microsoft labels "Anthropic models with Data Retention" are off by default in every region and need a separate opt-in. For those, Microsoft says Anthropic acts as an independent processor under Anthropic's own terms and stores most inputs and outputs for up to 30 days, or longer where its classifiers flag a possible policy violation. Enabling them is a new vendor decision and belongs in procurement and privacy review.

Copilot Chat usage limits and the licensing choice

Copilot Chat is the version of Copilot included with Microsoft 365 subscriptions for people who do not have the paid Microsoft 365 Copilot licence. Microsoft's partner announcement of 4 November 2026 says fair use limits now apply to more of its features, starting 2 November for organisations with 2,000 or fewer seats and 9 November for larger ones. It cites Message Center posts MC1536158 and MC1536163. Microsoft's wording is that "Admins can't view or adjust them, and Microsoft doesn't publish specific limits because they're subject to change", and that limits "can vary throughout the day".

When a user reaches a limit, a notice says when the feature will be available again, and a response already in progress is not interrupted. In most organisations the notice includes an option to request a licence, which an admin approves or denies in the admin center.

Unpublished limits that vary make capacity planning hard for teams that depend on the affected features. Licences should follow the work a role does, and a limit notice alone is weak evidence of need. The Register called the change a sales tactic, which is the reporter's opinion. Microsoft's own note does tell partners that the limits "create an upsell conversation", and that a partner report named "Users Experiencing Usage Limits" shows which customers have users reaching them. Review the licence request and self-service purchase settings before the first upgrade requests arrive.

Light users can stay on the included tier unless recurring limits interrupt necessary work. The sources cited here give no figures for how often users reach the limits, and they do not say that a paid licence removes every limit. Published numbers from Microsoft would change this advice, because the included tier could then be sized properly.

What is still unannounced or unclear

• The Learn page did not show the UK default change on 10 November 2026, and Microsoft has not documented how it determines that a tenant "already chose" a setting.
• The EU Data Boundary move has no date beyond "by the end of 2026", and in-country processing for Claude is not mentioned.
• The retirement date for the apps-only option is not public, and neither is the setting those tenants will be moved to.
• The Copilot Chat limits have no published numbers, and Microsoft's support page names only image generation as an example of a limited feature.

What to do this month

1. Open Copilot, Settings, View all, "AI providers operating as Microsoft subprocessors" and record the current Anthropic state, including whether the apps-only option is selected.
2. In a UK tenant, set Anthropic to off, to a named security group or to all users before 9 December, and record who approved the choice.
3. In an EU or EFTA tenant, keep the current setting until your Message Center confirms that Claude processing is inside the boundary, then review it.
4. Send the data protection officer the subprocessor details and ask whether the DPIA, the record of processing or any works agreement needs an update.
5. Brief the help desk on Copilot Chat limit notices, and decide whether users may request or buy Copilot licences themselves.

QueryNow helps IT teams review Copilot configuration and governance decisions like these through its Copilot AI Foundation service.