Anthropic Is a Microsoft Subprocessor: How to Check Your Microsoft Account

Someone on your team probably shared a screenshot. It said Claude is inside Microsoft 365 now. Then someone else said Microsoft is handing your company files to a different AI company. Both of those are part right and part wrong, and the wrong part is the part people repeat.

Here is the honest version up front. Microsoft's own help page says Anthropic, the company that makes Claude, has onboarded as a Microsoft subprocessor. The date to use is January 7, 2026. That is real, and Microsoft wrote it down in public. But it only covers some Microsoft products, not all of them. And if you run your business in Canada, it is probably already switched on in your account right now.

I am going to explain the whole thing in short words. Not because you are slow. Because this topic is buried in legal language on purpose, and you have a business to run.

The short answer

Microsoft's public documentation says Anthropic onboarded as a Microsoft subprocessor, effective January 7, 2026. That covers Microsoft Online Services, Microsoft Copilot, Copilot Studio, Power Platform and Copilot in Microsoft 365 apps. It does not cover Azure or Microsoft Foundry. In Canada, these models are turned on by default unless an admin turned them off.

What a subprocessor actually is

Think about building a house.

You own the house. You hire a builder. The builder does not do every job alone. He hires an electrician. The electrician works on your house, but he answers to the builder, follows the builder's rules, and the builder stays on the hook if the wiring is wrong. You never signed a separate deal with the electrician. You do not have to chase him. You chase the builder.

Data works the same way, with different words.

• You are the owner. The customer data, the emails, the files, they are yours. The legal word is controller.
• Microsoft is the builder. It handles your data on your behalf, under the contract you already signed. The legal word is processor.
• A subprocessor is the electrician. It is a company the processor brings in to do part of the work, under the processor's rules and the processor's contract.

So when Microsoft lists Anthropic as an AI subprocessor, it means this. Anthropic's models can do part of the work inside certain Microsoft products, and they do it under Microsoft's contract with you, not under a separate deal you have to go read and sign.

What changed on January 7, 2026, and why an owner should care

Microsoft's page, titled "Anthropic models in Microsoft Online Services," says it plainly. The exact words are: "To enable this change, Anthropic has onboarded as a Microsoft subprocessor. This change simplifies the experience and strengthens compliance and security under Microsoft's enterprise framework." Microsoft also told admins about it through Message Center post MC1193290.

What changed is who is responsible, not what the AI can do. Before, if you wanted Claude working inside Microsoft 365 Copilot, that ran under Anthropic's own commercial terms. Different company, different paperwork, different promises. Now, for those products, it runs under Microsoft's enterprise framework, the same one your other Microsoft services already sit under.

Why you should care, in one sentence: it means the AI part of your Microsoft stack is covered by an agreement you already have, instead of one you never read. That is genuinely simpler. It is also the exact reason nobody noticed it turn on.

One more detail a compliance officer will ask about, so know it now. There is a carve-out. For what Microsoft calls "Preview models with Data Retention," Microsoft's page says "Anthropic acts as an independent data processor, not a Microsoft subprocessor." Those preview options are off by default. If you never turn them on, the carve-out never applies to you.

Three things people say that are not true

1. "This has been true since September 2025"

No. September 24, 2025 was a different event. That is when Claude models became available inside Microsoft 365 Copilot under Anthropic's own commercial terms. That is close to the opposite of subprocessor status, because the terms were Anthropic's, not Microsoft's. If you are writing a policy note or answering a client, use January 7, 2026 for the subprocessor change.

2. "Anthropic is a Microsoft subprocessor everywhere, including Azure"

No. This is the trap that catches smart people. On Azure and Microsoft Foundry, Microsoft says the opposite. In Microsoft's own words there, Anthropic operates the inference and is the data processor and SLA provider. So on that side of the house, Anthropic is not a Microsoft subprocessor. Keep every subprocessor sentence you write scoped to Microsoft Online Services, Microsoft Copilot, Copilot Studio, Power Platform and Copilot in Microsoft 365 apps. Do not stretch it over to Azure.

3. "Anthropic announced it became a subprocessor"

No. Microsoft lists Anthropic as an AI subprocessor. Anthropic has never used the word publicly. That sounds like a small thing. It is not, because if you tell a client "Anthropic announced" and they go looking for the announcement, they will not find one, and now your whole memo looks shaky. Say what Microsoft says, and point at Microsoft's page.

A related caution. The Microsoft Online Services Subprocessor List itself sits behind a sign-in, so nobody outside can quote it directly. Do not write "Anthropic was added to the list on this date" as if you read the list. Write what Microsoft's public documentation says. That is the part you can back up.

The Canadian part, which is the part that matters today

Here is the practical bit. Microsoft enables Anthropic models on by default for most customers in commercial cloud. The exceptions are the EU, EFTA countries and the UK. Canada is commercial cloud. Canada is not on the exception list.

So read that again. If you are a Canadian business on Microsoft 365, this is very likely already on in your account, and nobody asked you. That is not a scandal, it is how default settings work. But you should know the switch exists and you should decide on purpose instead of by accident. Also worth logging: Anthropic models are currently excluded from the EU Data Boundary, which matters if you handle European data.

What to check in your own Microsoft account

This takes about four minutes. You need admin rights.

1. Sign in to the Microsoft 365 admin center.
2. Open Copilot.
3. Go to Settings.
4. Click View all.
5. Find AI providers.
6. Look at whether Anthropic models are on or off for your organization.
7. Decide. On is fine for most small businesses. Off is fine too. What is not fine is not knowing.
8. Write down what you chose, the date, and who chose it. If a client or an insurer asks later, that one note is the whole answer.
9. Leave the preview data-retention options alone unless you have read them and have a reason.

If your team is going to start using AI tools seriously after this, do the boring setup work once instead of ten times. Our guide on rolling AI out across a team covers who owns what and how to keep people from freelancing with company data. When you get to wiring the tools into the systems you already pay for, the integration side is where the real time savings show up.

Steal this prompt: ask Claude about your own setup

Reading about this is one thing. Asking questions about your own business is better. Paste
the prompt below into Claude, fill in the two blanks, and it will explain your situation in
plain language and then let you ask it anything you want.

It follows the five-part structure we use for every prompt here: role, context, task, format,
and questions before it starts. The same shape we break down in the
MD file guide.

ROLE: You are a plain-language explainer of how AI tools handle business data. You explain things the way you would to a smart person who does not work in technology. No jargon unless you define it in the same sentence.

CONTEXT: I run a small business. Here is my setup: [DESCRIBE WHAT YOU USE, for example "we are on Microsoft 365 Business Standard, about 12 staff, we use Copilot in Word and Outlook"]. What worries me most about AI and my data is: [YOUR ACTUAL WORRY].

Facts you should treat as correct, because I verified them:
- Microsoft's documentation states Anthropic onboarded as a Microsoft subprocessor, effective 7 January 2026, covering Microsoft Online Services, Copilot, Copilot Studio, Power Platform and Copilot in Microsoft 365 apps.
- This does NOT apply to Azure or Microsoft Foundry. There, Anthropic is an independent data processor, not a Microsoft subprocessor.
- There is a carve-out: for "Preview models with Data Retention", Anthropic acts as an independent data processor and is not a Microsoft subprocessor. Those are off by default.
- Microsoft turns Anthropic models ON by default for most customers in commercial cloud, excluding the EU, EFTA and the UK. Canada is commercial cloud.

TASK: Explain what all of this means for my specific business, in plain words. Tell me what is actually happening to my data, what I should check, and what genuinely does not matter.

FORMAT: Short paragraphs. No bullet lists longer than five items. Define any term the first time you use it. When you have finished explaining, give me ONE complete markdown file in a single code block that summarises my situation, what is switched on, and what I should check, ready to save as ai-data-notes.md and load into a Claude Project. Then invite me to ask you anything else about my setup, and mean it.

Ask me up to 5 questions before you start.

Then keep going. Ask it what happens to a document you upload. Ask it whether your admin
should turn something off. Ask it the question you were slightly embarrassed to ask a vendor.
That is the whole point of having it.

Frequently asked questions

Does this mean Microsoft is sending my files to Anthropic?

It means Anthropic models can do part of the work inside certain Microsoft products, under Microsoft's enterprise framework. Whether anything of yours actually goes through them depends on which features your people use. Check the AI providers setting in the Microsoft 365 admin center and decide on purpose.

Is my Azure setup covered by this?

No. On Azure and Microsoft Foundry, Microsoft says Anthropic operates the inference and is the data processor and SLA provider, not a Microsoft subprocessor. That is a different arrangement with different paperwork.

Do I have to do anything as a Canadian business?

Nothing is required. But Microsoft enables these models on by default in commercial cloud, and Canada is commercial cloud, so you should look at the setting and write down your decision and the date.

Can I turn it off?

Yes. It is an admin setting. Microsoft 365 admin center, then Copilot, then Settings, then View all, then AI providers.

Why do people keep saying September 2025?

Because something real happened then. On September 24, 2025 Claude models became available in Microsoft 365 Copilot under Anthropic's own commercial terms. That is a different arrangement from subprocessor status, and mixing the two dates is the most common mistake in circulation.

What about the preview models I keep seeing mentioned?

For Preview models with Data Retention, Microsoft's page says Anthropic acts as an independent data processor, not a Microsoft subprocessor. They are off by default. Leave them off unless you have a specific reason and have read the terms.

Where can I read the source myself?

Microsoft Learn, "Anthropic models in Microsoft Online Services." The Microsoft Online Services Subprocessor List itself requires a sign-in, so the public documentation page is the thing to cite.

Where this came from

Everything above comes from Microsoft and Anthropic's own published material, checked on
22 August 2026. If you want to read the originals:

• Microsoft Learn: Anthropic models in Microsoft Online Services - the primary source. This is the page that says, word for word, that Anthropic has onboarded as a Microsoft subprocessor, and it is where the carve-out for retention-enabled preview models is written down.
• Microsoft 365 Blog: Expanding model choice in Microsoft 365 Copilot - the September 2025 announcement people mistake for the subprocessor change. Worth reading precisely because it is the thing that gets confused.
• Anthropic: Microsoft, NVIDIA and Anthropic announce strategic partnerships - Anthropic's own account of the relationship. Note that Anthropic never uses the word subprocessor anywhere in it.

Microsoft also notified administrators through Message Center post MC1193290. That one sits
inside your own admin centre rather than on the public web, so check it there if you have
access.

One honest limit. The Microsoft Online Services Subprocessor List itself is behind a sign-in
and we could not read it directly. So this article quotes what Microsoft states publicly, not
the list. If you need the line item for an audit, pull it from the Service Trust Portal
yourself.