Claude in Microsoft Copilot moves into the EU Data Boundary by the end of 2026

On 5 November 2026 Microsoft announced that Claude models by Anthropic in Microsoft Copilot are coming into the EU Data Boundary. The post is by Charles Lamanna, Executive Vice President for Copilot, Agents, and Platform. He wrote that customers will get Claude and keep Microsoft's regional data commitments.

The EU Data Boundary is Microsoft's promise to store and process data for Microsoft 365 and its other online services inside Europe. It covers the European Union (EU) and the European Free Trade Association (EFTA). Eligible Microsoft 365 customers there get it in their contract, with nothing to opt into or buy.

Microsoft says EU and EFTA organisations will be able to use Claude across Copilot experiences by the end of 2026, including Copilot Cowork. Processing will happen inside the EU Data Boundary, where their data is already stored. Claude will join other models that already process data inside the boundary, including models Microsoft operates.

Claude models already power Copilot Chat, Copilot in Microsoft 365 apps, Copilot Cowork and Copilot Code. Microsoft says they are covered by its enterprise data protection. It also says customer data is not used to train generative AI foundation models, and admins decide who can use Claude.

Organisations already using Claude in Copilot need to take no action. Once the change is available, processing moves inside the boundary automatically.

Organisations that have not enabled Claude can turn it on at any time. In the EU and EFTA the setting is off by default. Admins can enable it in the Microsoft 365 admin center for everyone or for chosen users and groups.

For finer control, organisations can keep Microsoft 365 data at rest in their own country. In-country data residency is offered in 10 EU and EFTA countries. Microsoft says it will share more details on the Claude change closer to availability.

For an EU or EFTA buyer that kept Claude off because it sat outside the boundary, the end of 2026 is the target. Microsoft gives no exact date. Our reading is that such a buyer can plan a small pilot now and widen it once Microsoft confirms availability.

Our reading is that the move touches two privacy documents. The data protection impact assessment records the privacy risks of a project. The records of processing list what personal data you process and where. Re-check both once Microsoft confirms the date.

The announcement does not say whether Microsoft's in-country processing commitments will cover Claude, and the Learn page says Claude is excluded from them today. A buyer that needs data processed in one country should ask Microsoft before relying on the boundary move.

Microsoft's Learn page, updated on 4 November 2026, says Anthropic is onboarded as a Microsoft subprocessor. A subprocessor is a company that processes data on Microsoft's behalf, under Microsoft's contract. Microsoft's Product Terms and its Data Protection Addendum (DPA), the contract on how customer data is protected, then apply to most Claude models.

Claude runs in Microsoft Copilot, Researcher, Copilot Studio, Power Platform and Copilot in Microsoft 365 apps. Microsoft's Enterprise Data Protection covers that use. Its Customer Copyright Commitment also applies to Claude in the products it covers, such as Microsoft Copilot and Copilot Studio.

Microsoft turns Claude on by default for most customers in commercial cloud, outside the EU, EFTA and the UK. The Learn page says Claude in these products is excluded from the EU Data Boundary today, and from in-country processing commitments where those apply. Some features work only when Claude is enabled, so turning Anthropic off may remove them.

Customers inside the EU Data Boundary and customers in the UK have Claude off by default. Their setting, called AI providers operating as Microsoft subprocessors, shows "No users" until an admin chooses otherwise.

EU, EFTA and UK organisations that opted in earlier under Anthropic's own commercial terms and data processing agreement must opt in again. The toggle starts at Off.

On 3 April 2026 Microsoft added a setting for EU, EFTA and UK tenants called Copilot in Microsoft 365 apps with Anthropic models. It lets an admin make Anthropic the default model for Copilot in Microsoft 365 apps.

Since 22 July 2026, non-federal customers in GCC, Microsoft's Government Community Cloud, can use Claude through a setting. Claude is not available to federal GCC customers, in GCC High or DoD, or in other sovereign clouds.

The announcement does not name the UK, and neither source gives a date for any change to the UK default. Our reading is that UK buyers can decide on Claude now, under today's terms. Those terms make Anthropic a Microsoft subprocessor under the Product Terms and DPA.

Our reading is that most commercial tenants in the US and the Gulf already have Claude switched on by default. A buyer there can check who in the tenant has it, and whether anyone decided that.

US government buyers on GCC High or DoD do not have Claude at all. A team that works across commercial and government tenants may find Claude in one and missing from the other.

An EU, EFTA or UK organisation that opted in under Anthropic's own terms has to opt in again under the subprocessor setting. Our reading is that its users could lose Claude if nobody does.

The Learn page names a separate group of models labelled Anthropic models with Data Retention. Its examples are Claude Fable 5 and Claude Mythos 5. These models need Anthropic to retain data. The data Anthropic stores is not covered by your Microsoft Customer Agreement, including the Product Terms and DPA.

In every scenario, Anthropic models with Data Retention are off by default. A tenant admin has to opt in and accept Anthropic's Commercial Terms of Service and Data Protection Addendum. The admin also chooses which users or groups get them, and Anthropic is then an independent processor.

Anthropic stores most inputs and outputs for up to 30 days. Its trust and safety classifiers can flag a possible Usage Policy violation. Anthropic may then keep the content for up to two years, and the classification scores for up to seven. The Learn page says Anthropic does not use retained data for training without express permission.

Some organisations are eligible for Fable 5.1 with Anthropic as a Microsoft subprocessor, under the Product Terms and DPA. For them no extra terms apply, and Anthropic keeps no customer content, including prompts and responses. Microsoft says to ask your account team about eligibility.

Neither source says the data retention models are part of the EU Data Boundary move. Our reading is that they should stay off unless your legal team has reviewed and signed Anthropic's terms.

A buyer that wants a Fable-class model under Microsoft's own terms can ask its account team about the Fable 5.1 route first.