Can I run Claude inside Microsoft 365? Yes — here’s how

If you’ve been using Microsoft 365 Copilot for a while, you’ll know it as an OpenAI-powered assistant baked into Word, Excel and Teams. That’s changing. Microsoft has been quietly opening Copilot up to other frontier models, and one of the questions we’re hearing most often from clients right now is simple: can I use Claude, or another model, inside Copilot — and should I?

Can I actually use a different AI model inside Microsoft Copilot?

Yes. Microsoft has moved from a single-model platform to a multi-model one. Anthropic’s Claude models — currently Claude Sonnet 4 and Claude Opus 4.1 — now sit alongside OpenAI’s models across several parts of the Microsoft 365 Copilot suite, including the Researcher agent, Copilot Chat, Copilot Studio (for building custom agents), Agent Mode in Excel, and the AI layer in SharePoint.

This isn’t a bolt-on third-party plugin. Anthropic has been onboarded as a formal Microsoft subprocessor, meaning Claude usage through Microsoft 365 falls under Microsoft’s own Product Terms and Data Protection Addendum, rather than a separate Anthropic contract. Anthropic doesn’t use this data to train its models without explicit permission.

Where does Claude actually show up?

• Researcher agent — choose Claude Opus for in-depth research and reports, or let two models work together (one drafts, one critiques for accuracy and citation quality)
• Copilot Chat — Claude is now available as a model option across the full chat experience, not just Researcher
• Copilot Studio — build custom agents powered by Claude instead of (or alongside) GPT models
• Excel Agent Mode — use Claude to build formulas, spot errors and analyse data directly in a spreadsheet
• Copilot Cowork and SharePoint’s agentic layer — these run on Claude by default, with no model-switching option

Do I need to buy a separate Claude licence to get this?

Not necessarily — and this is the bit that surprises most clients. Claude access through Microsoft 365 Copilot is included within your existing Copilot licensing at no extra cost, rather than requiring a standalone Claude Team or Claude Enterprise subscription. For many organisations, especially those wanting a first step into using Claude’s capabilities for research or chatbot-style tasks, enabling it inside the Microsoft 365 environment you already manage is the more practical route.

That said, “included” doesn’t mean “on by default” everywhere. In the UK, EU and EFTA regions, Anthropic models are switched off by default and must be deliberately enabled by a Global Administrator in the Microsoft 365 admin centre. Government and sovereign cloud tenants don’t have access at all yet, as Anthropic hasn’t completed the relevant certifications for those environments.

What about data protection and where the data actually goes?

This is the question worth slowing down for. Claude models used through Microsoft 365 Copilot and Copilot Studio are currently excluded from the EU Data Boundary and equivalent in-country processing commitments — which is exactly why UK and EU tenants have it disabled by default. If data residency or sovereignty is a compliance requirement for your organisation, this needs checking against your specific obligations before you flip the switch, not after.

How should a business actually approach enabling this?

The mechanics of turning Claude on are simple. The governance around it is where the real work sits — and it’s the same discipline we’d apply to any Copilot rollout. We’ve covered the groundwork in detail in our post on Microsoft 365 Copilot best practices, guardrails and licensing: clean permissions, sensitivity labelling, and DLP policies before you give any AI model access to your organisation’s data.

A real-world example

We recently worked with a client who wanted to give staff access to Claude for chatbot and research use cases, but with a tighter, more locked-down approach than a standalone Claude licence would allow on its own. Rather than purchasing separate Claude Team or Enterprise licensing, we enabled Claude access through their existing Microsoft 365 environment instead — keeping the deployment inside the ecosystem where we could apply access controls, sensitivity labelling and DLP policies to reduce the risk of confidential information being shared inappropriately.

The phased plan looked like this:

• Assign Microsoft 365 Copilot licences to the agreed staff group
• Enable Claude access for those users specifically, rather than the whole tenant
• Restrict access through Microsoft 365 controls and review sensitivity labelling
• Apply DLP policies to help prevent data loss
• Test the configuration thoroughly before wider rollout
• Brief staff in line with the client’s own AI usage policy

For a later phase — if the client moves towards deeper integration, internal data access, or a more enterprise-grade governance model — we’d recommend stepping up DLP and compliance controls further, typically via the Microsoft Purview Suite add-on. It meant the client could move forward with a genuinely useful chatbot rollout immediately, without committing to a heavier licensing or governance model before they actually needed one.

Should you switch a model on just because you can?

Not automatically. More model choice is a genuine advantage — Claude, GPT and others each have different strengths, and being able to pick the right one for a research task, a spreadsheet, or a custom agent is useful. But every new model you enable is another thing that needs the same access controls, sensitivity labelling and staff guidance as the rest of your Copilot estate. The question isn’t just “can we turn this on”, it’s “have we prepared our environment so turning it on is actually safe”.