Claude Is Already Inside Your Copilot: What Investment Firms Should Switch On, Off and Teach

1. The Short Answer: Find Out Who Can Pick Claude Today

If your firm pays for Microsoft 365 Copilot and its tenant sits outside the EU, EFTA and UK, your licensed staff can pick Claude today, unless an admin changed the setting. Microsoft has switched Anthropic's models on by default for most commercial cloud customers since January 7, 2026.

Claude appears in Copilot Chat, Researcher, Copilot Studio and Edit with Copilot in Word, Excel and PowerPoint, and it runs Copilot Cowork. The switch came with a data decision your firm may never have made: Anthropic works as a Microsoft subprocessor, and its models sit outside Microsoft's EU Data Boundary.

The fix is short. Decide who may pick Claude and for which documents, scope the admin setting to match, keep models with data retention off until counsel has read Anthropic's terms, confirm your records cover it, and teach people the rule.

Choosing Claude in Copilot means choosing a model inside Microsoft's features. Anthropic's plugins, organization-wide skills and owner-controlled connectors belong to Claude's own apps, and section 6 compares the routes.

This guide is operational guidance, and your compliance officer and counsel own the interpretation.

None of it needs a project. It needs one owner, one admin session and a written decision.

2. Where Claude Shows Up in Microsoft 365 Copilot

Claude appears in two ways. Sometimes a user picks it, and sometimes it is the engine behind a feature.

• Picked by the user. Users can choose Claude in Copilot Chat, Researcher and Copilot Studio, and when editing with Copilot in Word, Excel and PowerPoint. Microsoft's subprocessor page says: "In capabilities such as Edit with Copilot in Microsoft 365 apps or Researcher, users can select Claude."
• Built in. Copilot Cowork and the Word, Excel and PowerPoint Agents in the Copilot app "exclusively use Anthropic's AI models," according to Microsoft Learn. Anyone who opens them is using Claude, whether or not they picked it.

Claude Opus 5.5 rolled out to Copilot on September 22, 2026 (Microsoft's announcement). Fable 5.1 is available only in Copilot Cowork and Copilot Studio.

When this guide was checked in October 2026, no Claude model picker was found in Copilot in Outlook or Teams.

A license decides who sees any of it. Microsoft's support page says: "You must have a Microsoft Copilot license to use model choice in the Microsoft Copilot app (desktop and web app only)." The free Copilot Chat selector offers only Auto, Quick response and Think deeper.

Copilot Cowork also runs on usage-based Copilot Credits, a cost on top of the license. Count both when someone asks what Claude in Copilot costs.

3. Who Has It Switched On by Default

The default depends on where your tenant is, and in Europe, on when it was created.

• Most commercial cloud tenants: on. In Microsoft's words, "Microsoft enables Anthropic models on by default for most customers in commercial cloud (excluding EU/EFTA and UK)." That has applied since January 7, 2026, when Anthropic began operating as a Microsoft subprocessor.
• EU, EFTA and UK tenants: off, with an exception. The main setting defaults to no users. A separate setting that covers only Word, Excel and PowerPoint is on by default for newer tenants, and Microsoft's page on it says: "This setting is on by default for tenants in the EU, EFTA, and UK that were created after March 25, 2026."
• Models with data retention: off everywhere. Anthropic models with data retention, such as Fable 5.0, "have additional controls in the Microsoft 365 admin center and are off by default for all scenarios".

For comparison, OpenAI-operated models have been on by default for eligible commercial customers since July 24, 2026 (Microsoft Learn).

So at a US firm where nobody opened the setting, licensed users can pick Claude. A London or Dublin firm whose main setting shows no users can still find Claude in Word, Excel and PowerPoint if its tenant was created after March 25, 2026.

Ask your Microsoft 365 admin which case is yours before anyone drafts a policy. Every later decision depends on the answer.

4. Where the Data Goes

When a user picks Claude, the prompt goes to Anthropic, working for Microsoft. Microsoft's subprocessor page puts the arrangement in one line: "To enable this change, Anthropic has onboarded as a Microsoft subprocessor."

Three facts follow from that:

• Outside the EU Data Boundary. Anthropic models "are currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments." OpenAI-operated models, by contrast, "are included in the EU Data Boundary".
• Different terms for retention models. For Anthropic models with data retention, Anthropic acts as an independent processor under its own terms. Microsoft's summary: "This means that Anthropic (not Microsoft) stores most inputs and outputs for up to 30 days before deleting them."
• The record lands in the mailbox. Copilot interactions are stored in each user's mailbox, under Microsoft Purview retention and eDiscovery.

Microsoft names Fable 5.0 as one example of a retention model. Check which models your admin center lists under those extra controls before anyone switches one on.

A firm with no EU footprint still has a vendor question. Prompts that may hold deal data now reach Anthropic under Microsoft's arrangement, so the Microsoft entry in your vendor file should record it.

UK and Irish firms also face a staff training question, covered in the UK and Ireland training guide.

Whether this data path fits your obligations to LPs, portfolio companies or regulators is counsel's call. Your job is to make sure someone made it on purpose.

5. The Admin Settings That Matter

The main switch sits in the Microsoft 365 admin center: Copilot, Settings, View all, AI providers operating as Microsoft subprocessors, Anthropic. Changing it takes the AI Administrator or Global Administrator role.

Microsoft's access page gives the step: "Under Choose who can access Anthropic models for Copilot and generative AI experiences, select your users or groups and Save." Access is set per provider rather than per model, and "Up to 999 combined user and group assignments are supported per AI provider."

Copilot Cowork is the open question. Microsoft's Cowork pages disagree on whether the provider scoping applies to it, so test it in your own tenant with a user outside the scoped group before you rely on it.

Fill in the Default column with what your tenant shows today, date it and file it. That page is the first piece of your evidence.

6. Claude in Copilot, Anthropic's Office Add-ins and Claude Enterprise

If you are looking for AI finance tools that work with Microsoft Office, Claude reaches Office by three routes. They differ in who processes the data, where the records sit and who holds the controls.

Claude inside Microsoft 365 Copilot

You buy from Microsoft. Microsoft 365 Copilot costs $30 per user per month paid yearly, or $31.50 monthly on an annual commitment. Copilot Business is normally $21, with an $18 first-year promotion between July 1 and December 31, 2026, for Microsoft 365 Business customers (Microsoft pricing).

Anthropic works as Microsoft's subprocessor, the records sit in each mailbox under Purview, and your Microsoft admin holds the controls. This route suits staff who already live in Word, Excel and PowerPoint.

Anthropic's own Office add-ins

Anthropic's documentation says: "Claude for Excel is generally available to Pro, Max, Team, and Enterprise plans." Claude for PowerPoint and Claude for Word use the same wording and became generally available on May 7, 2026. And "Claude for Outlook is currently in beta and available to Pro, Max, Team, and Enterprise plans" (Outlook page).

Compliance should note two details. Add-in data is deleted on Anthropic's back end within 30 days, and add-in activity "is not included in Enterprise audit logs". On a Pro or Max account the add-ins run under consumer terms, so allow them only on the firm's Team or Enterprise accounts.

Claude Team or Enterprise

You buy from Anthropic, and the features are Anthropic's: projects, Cowork with organization approval settings, organization-wide skills, plugins, and connectors an Owner controls. Team costs $25 per standard seat a month, or $20 billed annually; Enterprise costs $20 per seat per month billed annually plus all usage at API rates, with a 20-seat minimum when bought self-serve (Anthropic pricing).

Anthropic's Microsoft 365 connector works in the other direction, connecting Claude itself to your Microsoft 365 data. It "is available on all Claude plans: Free, Pro, Max, Team, and Enterprise," stays read-only unless an admin turns on write tools, and needs one-time consent from an Entra Global Admin (Anthropic Help Center).

Where the private equity plugin fits

Anthropic's private equity plugin lives on this third route. It installs in Claude through Customize, then Plugins, and its data connectors, twelve MCP servers including FactSet, PitchBook, S&P Global and Daloopa, sit in Anthropic's companion financial-analysis plugin. Its README warns: "MCP access may require a subscription or API key from the provider."

On the Microsoft side, the comparable route is an agent your firm builds in Copilot Studio, one of the places users can pick Claude. Either way, someone at your firm writes down the screening criteria, memo template and data rules; the Claude skills guide covers that work on the Claude side.

Copilot now includes Claude, so compare route by route. Drafting inside Office for licensed staff fits Copilot; deal work that needs a project per deal, a connected data room and firm-specific skills fits Claude Team or Enterprise.

7. What Compliance Should Decide This Quarter

Four decisions, each written down and dated:

1. The policy line. One sentence in the AI use policy naming who may select Claude in Copilot and for which kinds of documents. The admin setting then follows the sentence.
2. The data path. Add Anthropic, as Microsoft's subprocessor, to the Microsoft entry in your vendor file, with a note that Anthropic models sit outside the EU Data Boundary. Counsel decides whether that fits your obligations.
3. The retention models. Record whether models with data retention stay off. If one is ever switched on, Anthropic stores most inputs and outputs for up to 30 days under its own terms, so it needs its own review.
4. The records. Copilot interactions sit in each user's mailbox under Purview retention and eDiscovery. Confirm your retention policy covers them, then search for one interaction where a pilot user picked Claude, so you know the record exists before anyone asks for it.

If an LP's due diligence questionnaire asks how the firm governs AI, you want these four answers already on file.

Each decision fits on a page. Together they let you answer an LP, an examiner or your own board with a document instead of a memory.

8. What to Teach Staff

Training for Claude in Copilot is short and specific. It covers four things:

• Where Claude appears. Show the model choice in Researcher and in Edit with Copilot, and explain that Copilot Cowork and the Word, Excel and PowerPoint Agents use Anthropic's models whether or not anyone picks them.
• Which documents go where. Walk through the policy line with the firm's own examples: a CIM, a credit agreement, an LP letter, a portfolio company's management accounts. Say which route each one takes.
• When to pick Claude. Teach people to choose a model for a reason and to compare outputs on their own work, rather than trusting anyone's general claim about which model is better.
• How to check the output. Every number in a draft traces back to the source workbook or document before it reaches an IC memo or an LP. Picking a different model changes nothing about who signs.

Run it in an hour, on your own documents, for everyone with a Copilot license, and record who attended.

The Copilot training guide covers the wider Copilot curriculum. Add the Claude hour to it instead of running a separate course.

9. When to Move to Claude Enterprise

Claude in Copilot covers drafting and research inside Microsoft's apps. Move to Claude's own plans when the work needs Anthropic's features:

• Firm workflows built in. Organization-wide skills are available on Team and Enterprise, and owners manage plugins in Organization settings, Plugins and skills, where a private GitHub repository can sync as the organization marketplace.
• The data room connected under your rules. An Owner enables each connector, each person still authenticates, and Owners set tool permissions to Always allow, Needs approval or Blocked, organization-wide.
• Cowork with your approval settings. In Claude's Cowork, "Automatically approve" mode is on by default and "Always allow" for connector tools is off by default, and on Team and Enterprise those sit in the organization's settings.
• Enterprise controls. SCIM, custom roles, audit logs (180 days of events, not chat content), the Compliance API, custom retention, IP allowlisting and spend limits per user are Enterprise only.

The data terms change too. Anthropic's Commercial Terms say: "Anthropic may not train models on Customer Content from Services." Data is stored in the United States, and Anthropic offers no EU storage option for Team or Enterprise, so moving doesn't settle an EU residency question either.

Running both costs real money. Assume a 20-person firm that keeps Copilot and adds Claude Team standard seats on annual billing: $30 x 20 x 12 = $7,200 a year for Copilot and $20 x 20 x 12 = $4,800 for Claude, or $12,000 before any Copilot Credits.

So give each route a job. The plan comparison weighs Team against Enterprise, and the rollout cost guide budgets the rest.

10. Where to Start

If you do one thing this week, ask your Microsoft 365 admin to open the Anthropic setting and tell you who can pick Claude today. Then work in this order:

1. Find your default: commercial cloud, or EU, EFTA and UK, and for the second, whether the tenant was created after March 25, 2026.
2. Scope the setting to a named pilot group while compliance writes the policy line.
3. Confirm the retention models are off, and record the decision.
4. Test Copilot Cowork's scoping with a user outside the group.
5. Check the records with one test search in Purview for a Claude interaction.
6. Teach licensed users where Claude appears, which documents go where and how to check the output.
7. Settle the Claude Enterprise question workflow by workflow.

Then walk the table again every quarter, because the defaults keep moving: January 7, March 25 and July 24 each brought a change in 2026.