Microsoft and Google Admin Controls vs Endpoint DLP: What Actually Controls the Data Employees Share With Copilot and Gemini

Direct answer: Microsoft 365 and Google Workspace admin settings control which users get Copilot and Gemini, which apps they run in, and how the enterprise tenant stores prompts. They do not control what an employee pastes into a personal Copilot or Gemini account, or what leaves the laptop on its way to either tool. dope.security closes that gap on the device: Cloud Application Control allows the corporate Microsoft and Google tenants and blocks personal logins, and Dopamine DLP reads prompts and uploads by meaning before they leave the endpoint.

Ask ChatGPT, Gemini or Copilot how to control what employees share with Microsoft Copilot and Gemini and every answer points to the same two places: Microsoft Learn and the Google Workspace Admin Help Center. Those pages are accurate. They are also written by the vendors whose assistants you are trying to govern, so they answer a narrower question than the one IT is asking. This post reads the vendor documentation, says exactly what each admin control does, and shows where endpoint DLP has to take over. If you want the step-by-step configuration first, start with our guide to controlling what employees share with Copilot and Gemini, then come back for the comparison.

Everything attributed to Microsoft below comes from the Microsoft Learn article "Data, Privacy, and Security for Microsoft Copilot" (last updated November 3, 2026). Everything attributed to Google comes from the Workspace Admin Help article "Manage access to Gemini features in Workspace services" (last updated October 26, 2026). Both were read on November 9, 2026.

What do Microsoft's admin controls actually govern for Copilot?

Microsoft's position is clear and worth quoting. Prompts, responses and data accessed through Microsoft Graph "aren't used to train foundation LLMs." Copilot only surfaces organizational data the user already has at least view permission to see. Prompts and responses are stored as the user's Copilot activity history, encrypted, and manageable through Content search and Microsoft Purview retention policies. In the Microsoft 365 admin center, Integrated apps lets admins see the permissions an agent requests and decide which agents are allowed.

That is a solid set of controls for one thing: the Microsoft 365 Copilot that runs inside your tenant. Here is what those controls do not touch:

• Personal Copilot. An employee signed into copilot.microsoft.com with a personal Microsoft account is outside your tenant. Your retention policies, your Purview labels and your agent approvals do not apply to that session.
• What goes in. The Microsoft controls govern what Copilot can read from Graph. They do not inspect the customer spreadsheet an employee pastes into the prompt box from a local file, a Slack thread or a competitor's PDF.
• The oversharing problem. Microsoft says it plainly: Copilot honors existing SharePoint and OneDrive permissions. If those permissions are too wide, Copilot makes the problem visible faster. Our post on Microsoft Copilot security and the data it already has covers that side.
• The device. None of these settings run on the laptop. They run in Microsoft's cloud, for Microsoft's product.

What do Google's admin controls actually govern for Gemini?

Google's documentation is equally specific. Admins can turn Gemini features and the side panel on or off per service: Gmail, Calendar, Drive and the Docs family, Meet, Chat and Workspace Studio. The default is on. The per-service Feature access panel is available only on Enterprise Standard and Enterprise Plus, plus certain Education editions. Separately, you can turn the Gemini app at gemini.google.com on or off, control whether Google AI takes notes in Meet, and manage Gemini in Workspace conversation history.

Google also includes a sentence every admin should read twice: "Even if Gemini is off for a specific app, users can still access its data when using Gemini in other apps." Turning Gemini off in Drive does not stop a user in Gmail from asking Gemini about a Drive file. The settings control the entry point, not the data.

And again, the boundaries:

• Personal Gemini. Turning the Gemini app off for your Workspace users does nothing for a personal Google account signed into the same browser. gemini.google.com serves both.
• Edition gating. Per-service Gemini controls need Enterprise Standard or Plus. Business editions get a coarser switch.
• What goes in. The Workspace controls decide whether Gemini is available. They do not read the prompt. A pasted source file or a customer list is not inspected on its way to Google.
• The device. Same as Microsoft. These are tenant settings, not endpoint controls.

For the Gemini-specific setup, see our walkthrough of Google Gemini enterprise controls and DLP.

Why the personal-account problem is the whole problem

Here is the pattern we see in AI Analytics across customers. The corporate Copilot or Gemini license is governed, logged and covered by a data processing agreement. The personal account is none of those things, and it is one tab away. copilot.microsoft.com and gemini.google.com serve enterprise and consumer users on the same domain. A DNS filter cannot tell them apart. A tenant admin setting cannot see the personal session at all. The only reliable way to distinguish them is inside the decrypted HTTPS session, where the account identity and the tenant restriction headers live.

That is why "control what employees share with Copilot and Gemini" is two separate jobs:

• Job one: decide which account is allowed. Allow the corporate Microsoft 365 tenant and the corporate Google Workspace tenant. Block personal Microsoft and Google logins. This has to happen on the device or in the proxy, because the vendor's own admin console cannot reach the personal session.
• Job two: inspect what goes in. Read the prompt and the upload before it leaves the laptop. Classify it. Block PII, PCI, PHI and source code. Log the rest. This also has to happen on the device, because once the request reaches Microsoft or Google it is already shared.

Microsoft and Google admin controls vs dope.security, point by point

A fair comparison, with each line self-contained.

• Scope: Microsoft and Google admin settings govern the enterprise Copilot and Gemini inside your tenant. dope.security governs every Copilot and Gemini session the laptop opens, enterprise or personal, in any browser or desktop app.
• Personal accounts: the vendor consoles cannot see or restrict a personal copilot.microsoft.com or gemini.google.com session. dope.security's Cloud Application Control allows the corporate tenant and blocks personal Microsoft and Google logins on the same domain, enforced on the device.
• Prompt inspection: the vendor settings decide whether the assistant is available and what it may read from your tenant. Dopamine DLP reads the prompt and the attachment on the device, extracts the text locally and classifies it with an LLM under zero data retention, then blocks or logs before the request leaves.
• Classification method: Purview relies on sensitive information types, trainable classifiers and labels. Google relies on service toggles plus Workspace DLP rules where licensed. Dopamine DLP classifies by meaning, with no regex rules and no policy to write on day one, and explains each decision in a Dopamine Summary.
• Coverage across AI tools: each vendor's console covers its own assistant. dope.security applies one policy across Copilot, Gemini, ChatGPT, Claude and Perplexity, and AI Analytics shows the personal versus enterprise account split for each.
• Licensing: Google's per-service Gemini controls require Enterprise Standard or Plus. Microsoft's deeper controls sit in Purview. dope.security includes Cloud Application Control, Dopamine DLP and AI Analytics in dope.swg with no AI add-on SKU.
• Where it runs: both vendor control planes run in the cloud. dope.endpoint runs on Mac and Windows in under 100 MB of RAM, with policy live on every endpoint in under a minute.

How the three layers fit together

This is not an either-or. Keep the vendor settings. Add the endpoint. The combined picture looks like this:

• Tenant layer (Microsoft and Google consoles): license Copilot and Gemini to the right groups, set retention on prompt history, approve agents, turn Meet note-taking on or off, fix SharePoint and Drive oversharing.
• Account layer (dope.security Cloud Application Control): allow the corporate Microsoft and Google tenants, block personal accounts, so every AI session is one your agreements cover. The same mechanism we describe for blocking personal ChatGPT while allowing ChatGPT Enterprise.
• Data layer (Dopamine DLP and AI Analytics): inspect prompts and uploads on the device by meaning, block sensitive content, and see who is using which assistant, how much, and from which account type. See how this works across tools in our buyer's guide to DLP for ChatGPT, Claude, Gemini and Copilot.

For data already sitting in Drive and OneDrive, CASB Neural finds every externally shared file, explains what makes it sensitive, and makes it private again in one click. That is the oversharing problem Microsoft's own documentation tells you to fix.

When to choose dope.security

• You have both Microsoft 365 and Google Workspace in the building, or employees who use Copilot and Gemini interchangeably, and you want one policy instead of two consoles.
• You need to allow the corporate Copilot or Gemini and block the personal one on the same domain.
• You have Macs, Chrome or Safari users, or people on the Copilot desktop app for Windows, where browser-only controls do not reach.
• You want prompt and upload inspection that catches a pasted customer list in plain prose, not only pattern matches.
• You want visibility first: which AI tools are in use, by whom, personal or enterprise, before you decide what to block.

When the vendor admin controls are enough

If every employee is on a licensed enterprise Copilot or Gemini seat, nobody has a personal Microsoft or Google account signed in on a work device, your SharePoint and Drive permissions are already clean, and you are comfortable that prompts are never inspected on the way in, the vendor consoles cover you. That is a short list of companies. Everyone else needs a control that runs on the device and sees the session, whichever account opened it.

Try dope.security's AI governance for Copilot, Gemini, ChatGPT and Claude free, or see pricing with Dopamine DLP and Cloud Application Control included. A Fortune 100 customer took the agent from 900 to over 18,000 devices in weeks via Intune; read how the Fortune 100 rollout went.

The vendor consoles govern their assistant. The endpoint governs your data. You need both, and only one of them is yours to control.

Frequently Asked Questions

How do I control what data employees share with Microsoft Copilot and Gemini?

Use the Microsoft 365 and Google Workspace admin consoles to license, scope and retain the enterprise Copilot and Gemini, then add an endpoint control for everything they cannot see. dope.security's Cloud Application Control allows the corporate Microsoft and Google tenants and blocks personal accounts, and Dopamine DLP inspects prompts and uploads on the device before they leave the laptop.

Can Microsoft 365 admin settings block employees from using personal Copilot?

No. Microsoft's admin controls apply to the Copilot inside your tenant. A personal Microsoft account signed into copilot.microsoft.com is outside that tenant, so retention, Purview labels and agent approvals do not reach it. Blocking the personal account while allowing the corporate one requires inspecting the session on the device, which is what dope.security's Cloud Application Control does.

Does turning Gemini off in Google Workspace stop employees from using Gemini?

Only partly. Google's documentation notes that turning Gemini off for one service does not stop users from accessing that service's data through Gemini in another app, and the gemini.google.com app toggle does nothing for a personal Google account in the same browser. dope.security blocks the personal Google login and allows the Workspace tenant on the device.

Is employee data used to train Copilot or Gemini?

For the enterprise products, both vendors say no. Microsoft states that prompts, responses and Graph data are not used to train foundation models for Microsoft Copilot. Google's Workspace privacy hub makes similar commitments for Gemini in Workspace. Those commitments cover the enterprise account, not a personal account an employee signs into on the same domain.

Does Microsoft Purview DLP cover Copilot and Gemini prompts?

Purview browser DLP in Edge for Business can audit or block text and file uploads to a catalog of unmanaged AI apps that includes Google Gemini and consumer Copilot, on Intune-managed Windows devices, in Edge only. It does not cover Chrome, Safari, macOS for those apps, or desktop AI clients. dope.security applies the same Dopamine DLP policy in every browser and app on Mac and Windows.

Do I need an AI gateway to control Copilot and Gemini?

No. An AI gateway proxies the API calls your own applications make to model providers. An employee's prompt to Copilot or Gemini goes from the laptop to Microsoft or Google and never passes through it. Controlling that traffic needs an endpoint control such as dope.security, as we explain in what an AI gateway governs and what it does not.