The short answer: yes, conditionally. During the public preview running today, Anthropic Claude powers the full Copilot in SharePoint experience when admins enable Anthropic as a Microsoft sub-processor. From the mid-June 2026 opt-out rollout, Microsoft has confirmed Copilot in SharePoint runs on OpenAI's GPT-5.4 Reasoning model with no customer model choice. Both can be true at once because the preview and general availability use different model paths.
This is the reconciliation. Microsoft has published two pieces of guidance that read differently on first scan. Read together with the dates, they say the same thing: Anthropic for the preview, GPT-5.4 for general availability, both inside the same Microsoft 365 Copilot licence and governance frame.
The two sources that look like they conflict
The Microsoft Learn documentation for AI in SharePoint setup, current as of May 2026, says this:
AI in SharePoint's refreshed public preview uses an advanced reasoning model to support multi‑step planning and execution across SharePoint content. To access the full preview experience, tenant administrators should enable Anthropic as an AI sub‑processor for Microsoft Online Services in the Microsoft Admin Center. If Anthropic isn't enabled, AI in SharePoint will use a fallback reasoning model to maintain preview availability, and some advanced capabilities may vary depending on the model used.
Source: Get started with AI in SharePoint (preview)
Message Center notice MC1311968, published 15 May 2026, announcing the opt-out rollout, says this:
Copilot in SharePoint is built with model agility at its core. The team continuously evaluates and adopts the latest models. Currently, customer-controlled model selection is not available. At opt-out preview rollout time, Copilot in SharePoint will run on OpenAI's GPT-5.4 Reasoning model.
Source: MC1311968
Two Microsoft sources. One says Anthropic. The other says OpenAI's GPT-5.4. Both are right.
The reconciliation in one table
The conflict resolves once you separate preview from general availability and read the dates.
| State | When | Model running |
|---|---|---|
| Preview, Anthropic enabled | Now through mid-June 2026 | Anthropic Claude (advanced reasoning) |
| Preview, Anthropic not enabled | Now through mid-June 2026 | Microsoft fallback model (unnamed) |
| EU/UK preview, Anthropic disabled by default | Now through mid-June 2026 | Microsoft fallback model unless admin enables Anthropic |
| General availability rollout | Mid-June 2026 onwards | OpenAI GPT-5.4 Reasoning |
The Anthropic switch is a preview-time decision for Copilot in SharePoint specifically. At general availability, Microsoft's own model takes over for the SharePoint experience. The Anthropic sub-processor switch in the admin centre still has effect for other Microsoft 365 Copilot capabilities (the Word, Excel, and PowerPoint Agents began running on Anthropic from January 2026).
How to check what model your tenant is running
There is no in-product indicator that tells an end user which model is handling their chat. The signal lives in the admin centre.
Go to the Microsoft 365 admin centre, then Copilot, then Settings, then Data access. Under "AI providers operating as Microsoft sub-processors", if Anthropic is enabled for your tenant, the toggle is on. If it is off (the default for EU and UK tenants), Anthropic is not being used.
For Copilot in SharePoint during preview, this toggle determines whether the full experience runs on Anthropic Claude (toggle on) or on the Microsoft fallback model (toggle off). From the mid-June 2026 opt-out rollout, the toggle no longer affects Copilot in SharePoint specifically (which switches to GPT-5.4 regardless), but continues to affect the Word, Excel, and PowerPoint Agents and any other Copilot experiences that use Anthropic.
If you want to validate the experience difference yourself, run the same prompt in two tenants — one with Anthropic enabled, one without — and compare the response style. The output structure and verbosity differ noticeably.
What changes for your data when Anthropic is enabled
Anthropic operates as a Microsoft sub-processor under Microsoft's contractual safeguards. The high-level commitments:
- Coverage under the Microsoft Product Terms and the Microsoft Data Protection Addendum.
- Use of Anthropic models falls under Enterprise Data Protection and the Customer Copyright Commitment.
- Microsoft remains accountable for Anthropic's compliance under Microsoft's contractual commitments.
What it does not include:
- EU Data Boundary commitments. Anthropic is currently excluded from EU data boundary. This is the reason Anthropic is off by default for EU and UK tenants.
- In-country processing commitments, where applicable, also do not apply to Anthropic model traffic.
For most non-EU tenants, the practical effect is no change to existing data governance. The Microsoft frame still applies; Anthropic operates inside that frame.
For EU and UK tenants, the decision is real. Enabling Anthropic gives you the full preview experience but means some traffic falls outside EU data boundary commitments. Keeping Anthropic disabled keeps you inside the EU boundary but means you get the fallback model during preview (and GPT-5.4 anyway from mid-June, where the EU boundary question is different again).
For tenants with strict data-residency requirements where any cross-region processing is unacceptable, the safest path is to keep Anthropic disabled through preview and wait for general availability when Copilot in SharePoint runs on GPT-5.4 Reasoning.
The EU and UK situation in more detail
Microsoft has rolled Anthropic in gradually as a sub-processor across Microsoft 365 Copilot capabilities. From January 2026, Anthropic operates as a Microsoft sub-processor for Word, Excel, and PowerPoint Agents under contractual safeguards. Full availability across regions was expected by end of February 2026.
For EU and UK tenants specifically, Microsoft made a deliberate choice not to enable Anthropic by default because of the EU data boundary exclusion. The setting is opt-in, not opt-out, in those regions. This avoids putting traffic outside the EU boundary without explicit administrator consent.
For EU and UK tenants who want the full Copilot in SharePoint preview experience right now, the steps are:
- Microsoft 365 admin centre → Copilot → Settings
- Data access page → AI providers operating as Microsoft sub-processors
- Read the terms carefully (this matters more for EU/UK than elsewhere)
- Toggle "Enable Anthropic as a Microsoft sub-processor subject to the above terms"
The toggle applies across the Microsoft 365 Copilot stack, not just SharePoint. Once enabled, Anthropic models can be used in Word, Excel, PowerPoint Agents and any other capabilities that route through Anthropic.
If you are an EU or UK admin trying to make this decision before the mid-June rollout, the question to ask your data protection team is whether the preview experience between now and mid-June is valuable enough to justify the temporary boundary exclusion. The answer depends on your team's adoption posture. If you are using the preview to build SharePoint.md files and Skills before the rollout (recommended), the Anthropic-powered experience produces better output for that work. If you are waiting for general availability anyway, Anthropic during preview adds little.
Claude vs GPT inside Copilot, in practice
I have configured and run Copilot in SharePoint with and without Anthropic across client tenants since the September 2025 preview opened. Some observations on how the two models behave in this specific context.
Claude tends to be more verbose. A Skill output that takes 200 words on GPT often runs to 300-400 on Claude. The extra material is usually explanation, qualification, or context. For executive summaries, this is sometimes too much; for detailed analyses (contract reviews, audit reports), it is often welcome.
Claude follows multi-step instructions more reliably. When a Skill has six or seven steps with specific sub-rules per step, Claude is less likely to skip or merge steps. GPT is more likely to combine steps it thinks are similar.
GPT is more terse and direct. Where Claude might add a caveat or a qualifier, GPT will give the answer and stop. For one-off prompts where you want a quick result, GPT often lands better.
Both produce production-quality output. Neither is universally better. The Skills I run on Claude work on GPT and vice versa. The output style differs; the underlying capability is comparable.
None of this should drive a buy decision. The model selection for Copilot in SharePoint at general availability is GPT-5.4 with no customer choice. The question for the preview period is purely whether to opt into the better Anthropic-powered experience now or wait six weeks.
The wider context: Anthropic across Microsoft 365
Copilot in SharePoint is one part of a wider Anthropic-as-sub-processor pattern across Microsoft 365 Copilot. From January 2026:
- Word Agent, Excel Agent, and PowerPoint Agent run on Anthropic models when enabled
- Microsoft 365 Copilot Chat can use Anthropic for certain workloads
- Cowork (Microsoft's Anthropic-built agentic AI workspace) uses Anthropic natively
The Anthropic sub-processor toggle in the admin centre is a single switch that affects this whole stack. The decision to enable or disable Anthropic is therefore not a Copilot in SharePoint decision; it is a Microsoft 365 Copilot decision that has implications across all the capabilities mentioned above.
For Copilot in SharePoint specifically at general availability, the Anthropic decision becomes moot (GPT-5.4 takes over). For the other capabilities, the Anthropic decision continues to matter beyond mid-June 2026.
What to do this week
Three actions, depending on your tenant region.
Non-EU, non-UK tenant. Anthropic is likely already enabled by default. If you are running Copilot in SharePoint preview now, you are getting the Anthropic-powered experience. Nothing to do unless you want to verify in the admin centre.
EU or UK tenant where the data boundary is a hard requirement. Keep Anthropic disabled. Use the fallback model through preview. From mid-June 2026, Copilot in SharePoint runs on GPT-5.4 regardless, which puts you back inside the standard Microsoft model frame.
EU or UK tenant where you want the full preview experience. Review the terms with your data protection team, make the decision deliberately, and enable Anthropic if it lands the right way. Document the decision in your change-control record.
In all cases, the Anthropic decision is separate from the Copilot in SharePoint opt-out decision. The admin guide for the opt-out rollout covers the scope decision (do nothing, opt out tenant-wide, scope to selected sites). The Anthropic decision is an overlay on whichever scope option you pick.
For organisations rolling Copilot in SharePoint out across multiple teams, the Copilot in SharePoint Immersive is a one-day session where we work through both decisions (opt-in scope and Anthropic sub-processor) in your tenant. The day includes the model decision conversation specifically because the EU/UK tenants we run it with often need to make this call alongside the rollout.
