Since early 2026, something has quietly changed inside Microsoft 365 Copilot: when you use Copilot in Excel or PowerPoint, it may no longer be an OpenAI model answering your queries. It’s Claude, built by Anthropic. For IT teams and compliance officers, this shift raises a concrete question: does your Microsoft tenant data leave the Azure infrastructure?

The short answer is yes — and here’s what that means in practice.

Anthropic Is Now a Microsoft Subprocessor

As of January 7, 2026, Anthropic is officially listed as a subprocessor for Microsoft Online Services, including M365 Copilot and Copilot Studio. Since May 4, 2026, Claude is the default model in Copilot for Excel, PowerPoint, and Word (rolling out summer 2026).

Where Does Your Data Go?

Unlike OpenAI models — which run on Microsoft’s Azure — Claude models run on Anthropic’s servers (AWS/GCP, primarily in the US). Two direct consequences: (1) Excluded from the EU Data Boundary — Microsoft’s data residency commitments don’t apply. (2) No in-country processing — even in-country guarantees don’t extend to Claude.

Default Behavior by Region

To enable Claude for an EU/UK tenant: Admin Center → Copilot → Settings → View all → AI providers operating as Microsoft subprocessors. Global Administrator role required.

Contractual Protections

Microsoft’s DPA applies: Anthropic cannot use your data to train its models. The Microsoft Customer Copyright Commitment also applies. These are solid protections — but they don’t resolve the question of where, geographically, processing takes place.

Impact on Your Compliance Posture

For organizations subject to GDPR or data sovereignty policies, enabling Claude requires upfront analysis. Note: US tenants were opted in by default with no admin action required — a change that may have flown under the radar.

Sources: Microsoft Learn | Prosperity 247