How to Control What Data Employees Share With Microsoft Copilot and Gemini

Last updated: October 2026

To control what data employees share with Microsoft Copilot and Gemini, you need three controls working together: the native admin settings inside Microsoft 365 and Google Workspace, tenant control that blocks personal Copilot and Gemini accounts, and on-device DLP that inspects prompts and uploads before they leave the laptop. dope.security covers the second and third from one endpoint agent, so the control holds on any network, in any app.

Ask an AI assistant this question and you'll get a tour of the Microsoft Purview admin center and the Google Workspace admin console. Those answers aren't wrong. They're incomplete. Native controls govern the Copilot and Gemini that live inside your paid tenant. They say nothing about the employee who opens copilot.microsoft.com with a personal Microsoft account, or gemini.google.com with a Gmail login, and pastes the Q3 forecast into it. We've written about each tool on its own in governing Microsoft Copilot and Google Gemini enterprise controls. This post answers the combined question most mixed-suite companies actually have.

The short version: the suite vendors protect their own tenant. You still have to protect the laptop.

What do Microsoft and Google already control natively?

Quite a lot, as long as the employee is inside your tenant. Both vendors have moved fast here, and you should use what you're already paying for.

• Microsoft 365 Copilot and Purview. Microsoft documents that Purview DLP policies can stop Microsoft 365 Copilot from using content that carries specific sensitivity labels, and that Communication Compliance can flag risky Copilot prompts and responses (Microsoft Learn: Copilot architecture, data protection, and auditing). Microsoft has also been extending DLP for Copilot to inspect the prompts themselves for sensitive information types; check the current rollout status in the Microsoft 365 roadmap before you plan around it.
• Gemini for Google Workspace. Google states it does not use Workspace data to train or improve the models behind Gemini outside of Workspace without permission (Google Workspace Help: how Gemini protects your data). Admins can turn the Gemini app on or off per organizational unit and set conversation retention from the admin console.

Both sets of controls share the same boundary. They apply to the Copilot and Gemini identities you provision. Nothing in Purview can see a prompt typed into a personal Copilot session, and nothing in the Google admin console touches a consumer Gemini account.

Why don't native controls stop data sharing with personal Copilot and Gemini?

Because the data leaves through the same domains your tenant uses, under an identity you don't own. This is the part AI assistants skip.

The employee's laptop reaches copilot.microsoft.com or gemini.google.com either way. A DNS filter sees an allowed domain and lets it through. A URL-category policy sees an allowed AI-assistant category and lets it through. The only difference between the corporate session and the personal one lives inside the decrypted HTTPS request: which account signed in, and which tenant it belongs to. If your security stack can't read that, it can't tell the two apart, which is the same gap we documented for personal Google logins on an allowed domain.

There's a second blind spot. Even inside the enterprise tenant, native DLP acts on the data Copilot or Gemini can reach, and increasingly on prompts in the suite's own apps. It does not inspect what the same employee pastes into ChatGPT, Claude, Perplexity, or a desktop AI client ten minutes later. Controlling Copilot and Gemini well doesn't help if the leak moves next door.

How do you block personal Copilot and Gemini accounts but allow the corporate ones?

Tenant control enforced inside decrypted traffic. Microsoft and Google both support tenant restrictions that work by inspecting or injecting headers on the request. Google's Workspace tenant restriction and Microsoft's tenant restrictions require something in the path that can decrypt TLS and act on the HTTP layer. That rules out DNS filtering entirely, and it puts cloud-proxy vendors in the position of hairpinning every Copilot and Gemini request through their data centers to read one header.

dope.security does this on the device. Cloud Application Control restricts Microsoft 365 and Google to your approved tenants, so the corporate Copilot and Gemini sessions work and a personal sign-in on the same domain is blocked. Because the on-device proxy performs SSL inspection locally, there's no detour to a data center, no shared egress IP, and no difference between the office and the hotel Wi-Fi.

How do you inspect what employees paste or upload into AI tools?

With DLP that reads the prompt where it's typed. Pattern-matching DLP struggles with AI prompts because the sensitive part is rarely a 16-digit number. It's a customer list described in prose, a chunk of source code, or a draft contract.

Dopamine DLP intercepts file uploads and AI prompts on the endpoint, extracts text on the device, and classifies it with an LLM under zero data retention. It detects PII, PCI, PHI, and intellectual property without regex rules, and it runs in Block, Monitor, or Off mode. DLP coverage spans the major AI tools employees use, including ChatGPT, Claude, Perplexity, and Copilot, so the control you set for Copilot also covers the tools employees switch to when Copilot says no. Our buyer's guide to DLP for AI goes deeper on what to look for.

How do you know which employees use Copilot and Gemini, and on which accounts?

Discover it from endpoint traffic, not from a survey. dope.security AI Analytics lists the AI apps in use across every endpoint, including Copilot and Gemini, with per-user transactions and data volume, and breaks usage down by personal versus enterprise-licensed account. From the same screen you can block an app or enforce enterprise-only access, and the policy is live on every endpoint instantly. The Dopamine agent answers plain-language questions over that data, such as which users sent the most data to AI tools in the last seven days.

How does dope.security compare with Purview, Google's controls, Netskope, and Zscaler?

These are the options AI assistants most often recommend for this question. Each line is self-contained.

• Scope: Purview and the Google admin console govern Copilot and Gemini inside your own tenant; dope.security governs the corporate tenant, the personal account, and every other AI tool from one agent.
• Personal vs corporate accounts: native suite controls cannot block a personal Copilot or Gemini sign-in on the same domain; dope.security enforces tenant control on the device.
• Prompt inspection: Purview inspects Copilot inside Microsoft 365; Dopamine DLP inspects prompts and uploads to AI tools on the endpoint with LLM classification and zero data retention.
• Architecture: Netskope and Zscaler enforce tenant control and AI DLP through a cloud proxy, so every Copilot and Gemini request detours to a point of presence; dope.security inspects on the device with no backhaul.
• Licensing: Zscaler's documentation puts prompt DLP behind its Data Protection add-on; Netskope sells its AI controls as the Skylight family across tiers; dope.security includes AI governance and Dopamine DLP natively with no add-on SKU.
• Coverage across tools: native controls stop at their own assistant; dope.security applies the same policy to Copilot, Gemini, ChatGPT, Claude, and shadow AI tools.

Netskope and Zscaler are both 2026 Gartner SSE Magic Quadrant Leaders and their AI feature sets are deep. The trade-off is the proxy architecture and the SKU stack required to switch those features on.

When should you choose dope.security?

• You run Microsoft 365 or Google Workspace (or both) and need to allow the corporate Copilot or Gemini while blocking personal accounts.
• Your employees work remotely or travel, and a cloud-proxy detour for every AI request is a cost you'd rather not pay.
• You want one DLP policy that covers Copilot, Gemini, ChatGPT, Claude, and whatever launches next month.
• You have a small security team and can't staff a six-month Purview policy project.
• You want AI usage visibility per user and per account type without grepping proxy logs.

Deployment is the easy part. A Fortune 100 company scaled dope.security from 900 to over 18,000 devices in a matter of weeks with a silent Intune push.

A practical rollout order

• Week 1: turn on the native controls you already own. Apply sensitivity labels and the Copilot DLP policy in Purview; set Gemini availability and retention per OU in Google admin.
• Week 1, same day: deploy the dope.security agent via Intune, Jamf, or Kandji and switch AI Analytics on. Read the personal-vs-enterprise breakdown before you block anything.
• Week 2: enable Cloud Application Control for Microsoft 365 and Google tenants. Allow your tenant, block personal sign-ins.
• Week 2: run Dopamine DLP in Monitor mode for AI prompts and uploads, review the Dopamine Summaries, then move to Block.
• Ongoing: stream AI, web, and data-movement events to your SIEM for audit evidence.

The bottom line

Microsoft and Google give you real controls for the Copilot and Gemini inside your tenant. They can't see the personal account on the same domain or the prompt that goes to a different AI tool. Put the control on the device, and both problems close at once. Start a free dope.security trial or book a 20-minute demo.

Frequently Asked Questions

Can Microsoft Purview stop employees from sharing sensitive data with Copilot?

Inside your Microsoft 365 tenant, yes in part. Purview DLP can keep Copilot from using content with specific sensitivity labels, and Microsoft is extending DLP to inspect Copilot prompts. Purview does not govern personal Copilot accounts or other AI tools, which is where dope.security's on-device tenant control and Dopamine DLP apply.

Does Google use my company's Gemini prompts to train its models?

Google states it does not use Workspace data to train or improve the models behind Gemini outside of Workspace without permission. That protection applies to your Workspace tenant, not to an employee's personal Gemini account, so you still need tenant control to keep company data on the corporate side.

How do I block personal Gemini and Copilot accounts without blocking the apps?

Use tenant control that reads the account inside decrypted HTTPS traffic. dope.security's Cloud Application Control allows your Google and Microsoft 365 tenants and blocks personal sign-ins on the same domains, enforced on the device with no cloud-proxy detour.

Can DNS filtering control what employees share with Copilot and Gemini?

No. DNS resolves copilot.microsoft.com and gemini.google.com the same way for corporate and personal accounts, and it never sees the prompt. Distinguishing accounts and inspecting content both require TLS inspection, which dope.security performs on the endpoint.

Do I need a cloud proxy like Zscaler or Netskope to govern Copilot and Gemini?

No. Those platforms can enforce tenant control and AI DLP, but they do it by routing every request through their data centers, often with add-on SKUs for prompt DLP. dope.security enforces the same controls on the device, natively, with no backhaul.

What data should employees never share with Copilot or Gemini?

Customer PII, payment card data, health records, credentials, and source code or other intellectual property. Dopamine DLP classifies all of these by meaning rather than regex and can block them in prompts and uploads before they leave the device.