What changed, exactly
On 8 December 2025 a new admin setting appeared in the Microsoft 365 admin center titled “AI providers operating as Microsoft subprocessors”. Anthropic was the first vendor exposed under it. On 7 January 2026 the legacy admin toggle for opting in to Anthropic under Anthropic's separate commercial terms and data processing agreement was deprecated, and Anthropic was onboarded as a Microsoft subprocessor under Microsoft's enterprise framework instead.
For most commercial-cloud customers the new toggle is set to ON by default. The practical effect is that users in those tenants can pick Claude models inside Microsoft 365 Copilot, Researcher, Copilot Studio, Power Platform, Agent Mode in Excel, and the Word, Excel, and PowerPoint agents. UI indicators inside Copilot show when a Claude model is being used; in Copilot Studio the model has to be chosen at agent creation.
Microsoft expected full availability across all in-scope tenants by the end of March 2026. On 3 April 2026 a separate admin center setting was introduced that allows EU/EFTA/UK admins to enable Anthropic as the default Copilot model for M365 apps if they opt in - a second knob, distinct from the underlying subprocessor toggle.
Who is affected, in plain terms
- -Commercial-cloud customers outside EU/EFTA/UK: Anthropic ON by default. Your tenant is processing user prompts through Claude unless an admin disables the toggle.
- -EU / EFTA / UK customers: Anthropic OFF by default. The toggle is exposed but you have to actively opt in to expose Claude inside Copilot. If you previously opted in under the legacy Anthropic terms, you must opt in again under the new subprocessor toggle.
- -Government clouds (GCC, GCC High, DoD) and other sovereign clouds: Not in scope at all. No toggle exposed. FedRAMP certification has not been established.
- -Downstream B2B SaaS that uses Microsoft 365 to process customer data: Anthropic is now indirectly a sub-processor of YOUR data chain wherever Copilot touches the data. Your customers are entitled to know.
What Article 28(2) requires of you
Under GDPR Article 28(2), a processor must not engage a new sub-processor without the controller's prior authorisation, either specific or general. Commercial SaaS contracts almost universally operate under general written authorisation: the controller agrees up front that the processor may use sub-processors, subject to maintaining a current list and providing prior notice of additions or replacements so the controller can object before the change takes effect.
Microsoft introducing Anthropic as one of its own subprocessors cascades into your chain whenever you process personal data through Microsoft 365. The change is documented and dated; the obligation is on you to (a) update your published sub-processor list and (b) notify your customers in line with the notice window your DPA specifies, which is most commonly 10 to 30 days. The EDPB Opinion 22/2024 spells out that the identity of all processors and sub-processors must be “readily available at all times” - the legal expectation is a current, accessible list, not an annual refresh.
A customer notification template you can send today
Below is a starting-point Article 28(2) notification that names the actual change, cites Microsoft's documentation, and lays out the standard objection mechanic. Edit the bracketed fields, adapt the tenant-configuration paragraph to what you actually did, and send. The references at the bottom give your customers the same primary sources to verify against.
Subject: Sub-processor update: Anthropic via Microsoft 365 (Article 28(2) notice)
Hello {customer name},
This is a sub-processor notification under Article 28(2) of the GDPR and section {DPA section reference} of our Data Processing Addendum.
Effective 7 January 2026, Microsoft has added Anthropic, PBC as a sub-processor to Microsoft Online Services. Microsoft has enabled Anthropic models on by default for most customers in its commercial cloud (this excludes customers in the European Union, the European Free Trade Association and the United Kingdom, for whom Anthropic is off by default). Because {your company} uses Microsoft 365 Copilot as part of how we deliver {product/service}, Anthropic is now indirectly a sub-processor of {your company} where Copilot processes data on your behalf.
What Microsoft has said about scope:
- Anthropic models are excluded from the EU Data Boundary and from in-country processing commitments where applicable.
- Anthropic models operate under Microsoft's Product Terms and the Microsoft Data Protection Addendum.
- Government cloud (GCC, GCC High, DoD) tenants are not affected.
What this means for you:
- Our published sub-processor list now reflects this change (see {link to your subprocessor page}).
- We have {disabled Anthropic at the tenant level / restricted Anthropic to {team} / accepted the default and documented it}, in line with our overall data-processing posture.
- This notice is provided in compliance with the {N}-day prior-notice window under our DPA. If you wish to object, please reply to this email by {deadline}.
If you have any questions about how Anthropic is or is not used inside your data, please reach out to {DPO email or contact}.
{Signature, role, company}
References:
- Microsoft Learn, "Anthropic as a subprocessor for Microsoft Online Services":
https://learn.microsoft.com/en-us/microsoft-365/copilot/connect-to-ai-subprocessor
- Microsoft Online Services subprocessor list: https://aka.ms/subprocessorThe recurring problem this is one instance of
Microsoft + Anthropic is the highest-profile sub-processor change of 2026 so far, but it is not the only one and it will not be the last. Across the 18 SaaS providers Registora monitors daily, every one of them updates its own sub-processor list at least once a year. Stripe, AWS, Vercel, Cloudflare, Twilio, Sentry, OpenAI, Anthropic, and the others publish those changes to their own pages; the responsibility for cascading the relevant change into your own sub-processor disclosure and your customer notice sits with you. That is the recurring Article 28(2) work nobody wants to remember to run.
Registora hosts your subprocessor register on your own subdomain (or custom domain), monitors every upstream provider daily, amends your register the moment any of them changes, and drafts the Article 28(2) customer email for you to review and send. The free tier hosts up to five subprocessors. Growth (49 USD per month) covers the full list, custom domain, customer notification automation, REST API, and webhooks. See the chain report for the broader picture of how concentrated your upstream actually is, and /check to audit any public site for the third parties it loads.
Sources
- -Microsoft Learn - Anthropic as a subprocessor for Microsoft Online Services (canonical Microsoft documentation; covers the toggle, the dates, and the EU Data Boundary exclusion).
- -Microsoft Online Services subprocessor list (Service Trust Portal) - the live Microsoft sub-processor register.
- -Microsoft Products and Services Data Protection Addendum (DPA) - the contractual framework Microsoft cites as governing Anthropic use under its enterprise umbrella.
- -EDPB Opinion 22/2024 - sub-processor identities must be “readily available at all times”.
- -EDPB CEF 2026 - the 25-regulator coordinated enforcement action on transparency that puts Article 13/14 (and Article 28 by extension) under year-long supervisory focus.