The short answer

For a regulated firm, the setup matters more than the tool.

Copilot can already run Claude in Microsoft 365, and a firm can also use Claude on its own through a Team or Enterprise plan. Both can carry client work. Both need the same groundwork: limit what the AI can read, choose each group's models and keep a record. Get each client's consent before return information goes in. Copilot keeps Microsoft's labels and rules in the same system as your files. Claude on its own gets Anthropic's newest features first and adds a second vendor contract to manage. Either way, your administrator should limit who can open which files, set each group's AI models and note when each was set.

The comparison

The table has three columns because "Copilot" means two different things: the product as it arrives, and the same product once someone has set it up.

The first two columns show Copilot before and after setup. The third shows a Claude Team or Enterprise plan as it arrives. A Claude plan can be set up too: the same permission review and a written decision on the connector, and on Enterprise a retention period and the audit logs.

QuestionCopilot (as it arrives)Copilot (governed)Claude (Team or Enterprise)Edge
What it can readEverything the user can open, overshared files includedOnly what the permission review leaves openEverything the user can open, through the connectorDepends on setup
Which AI modelsMicrosoft's, OpenAI's and Anthropic's. The last two are on by default for most commercial customersMicrosoft's only for staff who handle returns, set group by group, with the date recordedClaude models, by planDepends on setup
Where prompts are processedNearest region, others at peak load. Outside providers have no in-country commitmentMicrosoft's regions, outside providers off. U.S.-only processing is expected by the end of 2026, not committed todayU.S.-only processing on usage-based Enterprise, at extra cost. Not on TeamDepends on plan
ContractMicrosoft's terms, except opt-in "Data Retention" modelsMicrosoft's terms, with a dated record of allowed modelsAnthropic's terms, a second vendorGoverned
Record of AI activityAvailable in Purview, once set upRetention switched on, then searched once to prove it worksCompliance API on the Enterprise planGoverned
Labels and rulesAvailable, but not set to the firm's filesLabels on the client folders, and rules that keep Copilot off them, testedCopilot's Purview rules do not apply. Check which Purview controls cover third-party AI appsGoverned
Proof for an examiner or insurerNone by defaultA dated record of every safeguardNone by default. A dated record once someone sets it upDepends on setup
Training on firm dataNot used for trainingNot used for trainingNot used by defaultTie
AgentsAgent Builder, included with the licenseAgents reviewed before use, each reading one libraryProjects and skillsDepends on setup
Where people use itWord, Excel, Outlook, Teams, PowerPoint and Copilot ChatWord, Excel, Outlook, Teams, PowerPoint and Copilot ChatAnthropic's app, reaching Microsoft 365 through a connectorBoth Copilots
Writing and analysisStrong, on any of the three model sourcesStrong. Staff who handle returns use Microsoft’s models, everyone else can use Claude if the firm allows itStrong, on Claude modelsTie
Anthropic's newest featuresOn Microsoft's scheduleOn Microsoft's scheduleAs Anthropic releases themClaude
Checked against Microsoft's and Anthropic's own documentation. The detail and every source are further down the page.

Where a prompt goes

Copilot (as it arrives)

  • Reads everything each person can open
  • Both outside AI providers on by default
  • No record
Copilot as it arrivesCopilot reads everything the person can open, including labeled and overshared files, and can send work to GPT models run by Microsoft, GPT models run by OpenAI, and Claude models run by Anthropic, with both outside providers on by default. There is no record.ReadsModelsEverything thisperson can openCopilot(as it arrives)GPTrun by OpenAIGPTrun by MicrosoftClauderun by AnthropicOutside providerson by defaultRecordNone

Copilot (governed)

  • Reads only what the permission review leaves open
  • For return work, models set on purpose and the reason written down
  • Every setting dated
Governed CopilotCopilot reads only what the permission review leaves open, labeled client files are kept out, the two outside model providers are switched off for anyone who handles return data, and every setting is recorded with a date.ReadsModelsWhat the permissionreview leaves openCopilot(governed)GPTrun by OpenAIGPTrun by MicrosoftClauderun by AnthropicOutside providersoff for anyone whohandles return dataRecordDatedevery setting, every review

Five questions before you buy another license

Before your firm assigns another Copilot license, you should be able to answer five questions in writing about every AI tool you use.

  1. What can it read?
  2. Which models run, and for whom?
  3. Where is the data processed?
  4. Under which contract?
  5. Where is the record of what the AI did, and when did someone last check it works?

If any answer is "we are not sure," that's the gap a governed setup closes.

The detail

The details

01

Your firm may already be using Claude

+

Most comparisons of Copilot and Claude look at the models, asking which one writes better or handles a workbook more cleanly. For a firm that holds client data, that isn't the choice anymore, because Copilot can already run Claude. For most commercial customers outside the EU, EFTA and UK, Microsoft turns it on by default.

Since July 9, 2026, models that OpenAI runs itself can also power Copilot's GPT-based features, and since July 24 they've been on by default for eligible commercial customers. So the question to ask is whether anyone at your firm decided what the AI can read and which models it may use. If someone did, where is that written down?

Copilot now uses models from three places. Two of them offer GPT models that OpenAI builds, so what sets the three apart is who runs each model.

GPT models Microsoft runs. OpenAI builds them, and Microsoft runs them in its own cloud under your firm's Microsoft agreement. They're what most people have in mind when they think of Copilot.

GPT models OpenAI runs. Microsoft now also offers OpenAI models that OpenAI runs itself. In that case OpenAI is a Microsoft subprocessor, meaning a company Microsoft hires to do part of the work. Microsoft says these models "serve the same GPT-based Copilot experiences your users already have." As of July 24, 2026, they're on for all users at eligible commercial customers unless an administrator turns them off.

Claude models Anthropic runs. Anthropic is a Microsoft subprocessor too, and Microsoft has Anthropic's models "on by default for most customers in commercial cloud (excluding EU/EFTA and UK)." That covers Copilot, Researcher, Copilot Studio, Power Platform and Copilot in the Microsoft 365 apps. Copilot Cowork, generally available since June 16, 2026, also runs on Anthropic models, but it stays off until an administrator turns it on.

Your administrator controls both outside providers in one place: the Microsoft 365 admin center, under Copilot, then Settings, then View all, then AI providers operating as Microsoft subprocessors. The administrator can apply it to specific users or security groups, so you don't have to choose between everyone and no one.

One group of Claude models needs a decision of its own. Microsoft's Product Terms and Data Protection Addendum cover Anthropic models in Copilot unless the models are labeled "Anthropic models with Data Retention." Microsoft keeps those models off until an administrator opts in. Once they're on, Anthropic acts as an independent processor under its own Commercial Terms of Service and Data Protection Addendum.

Microsoft says Anthropic "(not Microsoft) stores most inputs and outputs for up to 30 days before deleting them." Content flagged by Anthropic's safety classifiers can be kept for up to two years. If your firm opts in, record who decided and when.

Microsoft now calls its product for work accounts Microsoft Copilot, though until recently it was Microsoft 365 Copilot. The same Copilot app now serves personal accounts too, so you can no longer tell from the name alone which terms apply, only from the account. Microsoft says the organization's controls "continue to apply when you use Copilot with your work or school account."

Copilot also isn't ChatGPT, which is OpenAI's own app with its own accounts and its own terms. Say two staff members are working on the same return, and one pastes it into a personal ChatGPT account while the other asks Copilot about it. The one asking Copilot works under your firm's Microsoft agreement. The one using ChatGPT doesn't, even when the model family is the same.

02

What it can read

+

Copilot works inside the tools your firm already uses. It drafts in Word, works through a workbook in Excel, summarizes a thread in Outlook and catches someone up on a Teams meeting. Claude on its own works in Anthropic's app and reaches your firm's files through Anthropic's Microsoft 365 connector, which reads SharePoint, OneDrive, Outlook and Teams.

Both of them read what the user can already open, which most comparisons leave out. Microsoft says Copilot presents "only data that each individual can access." Anthropic says its connector works through delegated permissions, meaning it acts as the signed-in person, inside that person's existing Microsoft 365 access.

That would be fine if someone had set those permissions on purpose. In a small firm, usually nobody did, and they simply piled up.

Picture a compensation file in a folder eleven people can open, where the only thing protecting it was that nobody knew where it was. Either assistant finds it with one plain question. We covered where those permissions come from in Copilot Does Not Leak. It Reveals.

So the permission review, a check of who can open which files, isn't only for Copilot. It's the first step for any AI that reads your firm's files, Claude included.

Microsoft Entra is where Microsoft 365 keeps its sign-in and access settings. Anthropic's setup asks a Microsoft Entra Global Administrator to grant consent for the connector once, for your firm's whole Microsoft 365. That consent is a firm decision, so record which role gave it and when.

03

What Microsoft 365 protects that the Claude app cannot

+

Microsoft 365 has protections that Anthropic's app doesn't reach. Microsoft ships some default settings, but none of them know which folder holds your firm's returns.

Sensitivity labels. These are Microsoft's tags that mark how sensitive a file is, and a label such as Client Confidential can encrypt the file. Microsoft says that when content is encrypted this way, Copilot "honors the usage rights granted to the user." In other words, Copilot can do with the file only what that person is allowed to do. A label built around your firm's work puts that protection on the file itself, which a folder name can't do.

Rules for Copilot. Microsoft Purview, the data protection settings in Microsoft 365, can keep Copilot from using files and emails that carry sensitivity labels. It can also stop Copilot from answering when a prompt contains a sensitive number, such as a Social Security number. Microsoft still marks that second rule as preview.

A record of AI activity. Purview can search and keep Copilot prompts and responses. Anthropic's Enterprise plan offers a Compliance API that lets your systems pull activity feeds, chat data and audit log events. Neither one counts as a record until your administrator has set it up and checked that it runs.

Governed Copilot pulls ahead here, because the labels and rules sit in the same system as the files they protect, and one administrator sets all of them.

04

Where the data is stored, and where it is processed

+

These are two questions, and the answers are different. Where the data is stored is settled, and you can check it. Microsoft stores each Copilot prompt and response in the user’s own Exchange Online mailbox. That mailbox sits in your firm’s default geography, the region your Microsoft 365 is set up in. Under the Product Terms, Microsoft commits to keeping that stored content there when the default geography is the United States, or one of fourteen other countries. That commitment is why Purview can search and retain it.

Where the data is processed, meaning where the AI runs, is the open question, and for a tax firm it's the one that matters most. For the models Microsoft runs itself, it depends on how busy the data centers are. Microsoft routes Copilot requests for them "to the closest data centers in the region, but also can call into other regions where capacity is available during high utilization periods."

Microsoft has announced in-country processing for Copilot and expects to offer it for the United States by the end of 2026. Today it isn't a commitment for U.S. tenants.

The two outside providers are further from those commitments. Microsoft says Anthropic models "are currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments." It says OpenAI's own models "are currently excluded from in-country processing commitments when applicable."

Claude on its own isn't any simpler. For its API, the service software developers build on, Anthropic documents a U.S.-only setting for where the AI runs. The default there is global routing, though, which may run the AI in any available geography. On that platform, stored data is held in the United States. Anthropic offers U.S.-only processing only on its usage-based Enterprise plans, at extra cost. Team plans don't have it.

For a tax firm, this matters because of 26 U.S.C. §7216(a). It covers anyone in the business of preparing returns, or providing services in connection with preparing them. It makes it a misdemeanor for them to knowingly or recklessly disclose information furnished for a return, or to use it for any purpose other than preparing the return.

The exception in 26 CFR 301.7216-2(d)(1) lets a preparer share return information without the client's consent only with another tax return preparer located in the United States. It covers only sharing to prepare or help prepare the return, and only if the service is not a substantive determination. A provider of auxiliary services, and its employees who assist, count as tax return preparers.

26 CFR 301.7216-2(d)(1)

Except as limited in paragraph (d)(2) of this section, an officer, employee, or member of a tax return preparer may disclose tax return information of a taxpayer to another tax return preparer (other than an officer, employee, or member of the same tax return preparer) located in the United States (including any territory or possession of the United States) for the purpose of preparing or assisting in preparing a tax return, or obtaining or providing auxiliary services in connection with the preparation of any tax return, so long as the services provided are not substantive determinations or advice affecting the tax liability reported by taxpayers.

26 CFR 301.7216-2(d)(1), first sentence

The same paragraph closes by requiring the client's consent before return information goes to another preparer for substantive determinations. That's one reason a well-built AI workflow reports what doesn't match and leaves the judgment to the professional.

The regulation's own example treats a contractor's employee abroad who only views return information held on a U.S. server as a disclosure outside the United States. So for this rule, location means where the people who receive or view the information are, not where the provider is headquartered.

Microsoft commits to store Microsoft 365 Copilot prompts and responses at rest in the United States for U.S. tenants. Its data protection terms let it process data in the United States or any other country where Microsoft or its subprocessors operate. U.S.-only processing for Copilot is expected by the end of 2026, not committed.

Claremont's position is that client tax return information goes into Microsoft 365 Copilot only with each client's consent under 26 CFR 301.7216-3. That holds until Microsoft commits to U.S.-only processing and your firm can show who reaches the data. For Form 1040 clients that consent follows Rev. Proc. 2013-14, and the Social Security number stays out even with consent. Whether a particular disclosure needs consent is a conclusion for your firm's counsel.

Whether automated processing on a server abroad, with no person viewing it, is a disclosure hasn't been decided. Claremont doesn't build on the argument that it isn't. Either way, the settings step for everyone who handles return data is the same. Keep them on the models Microsoft runs itself, turn the outside providers off and record the date.

Patient data is its own case. Microsoft's page on Anthropic models in Copilot says nothing about HIPAA. Anthropic offers a business associate agreement (BAA) on its HIPAA-ready Enterprise plan, and the agreement covers chat. But Anthropic also says its own Cowork product "isn't an Eligible Service under the BAA in any configuration." Anthropic's page doesn't address Microsoft's Copilot Cowork, so a practice should keep patient information out of both until coverage is confirmed in writing.

05

When Claude on its own is the right choice

+

Claude on its own fits a regulated firm when it is set up on purpose. That means a Team or Enterprise plan under Anthropic's commercial terms, with the connector limited to named people and read-only unless someone decides otherwise. On Enterprise, it also means a retention period set and the audit logs in use.

It fits most easily when the firm doesn't run on Microsoft 365, when a team works without client data, or when the firm prefers Claude for writing and analysis and is ready to manage a second vendor contract. For client tax return information the rule is the same as for Copilot: each client's consent first.

Many firms land in between, with Copilot for everyone and Claude inside Copilot allowed for a named group that never handles return data. Whichever you choose, the same five questions apply. What can it read, and which models run? Where is the data processed, and under which contract? Where is the record of what the AI did?

Sources

§Sources16 sources · Microsoft 9 · Anthropic 5 · statute and regulation 2+

Claremont Security helps regulated firms get more done with AI on client work, with the setup and the record to show for it, whichever tool the firm uses. To see where a prompt goes before and after, try the AI Exposure Check. It's free and needs no account: claremontsecurity.com/ai-exposure

Get a one-page estimate for your firm: what the work would take and what it’s worth, from your own numbers. No charge.