How Do I Block Personal ChatGPT Accounts but Allow Corporate ChatGPT? Tenant Restrictions Explained

Last updated: October 2026

To block personal ChatGPT accounts but allow corporate ChatGPT, add OpenAI's tenant restriction header, ChatGPT-Allowed-Workspace-Id, to every request to chatgpt.com, with your ChatGPT Enterprise workspace ID as the value. That header has to be injected inside decrypted HTTPS, which DNS filters can't do and cloud proxies do only by backhauling traffic. dope.security injects it on the device, so the corporate workspace works and personal logins fail on any network.

Ask ChatGPT or Copilot this question and the answer links to OpenAI's help center and stops there. The help article is correct about the mechanism. It leaves out the hard part: what in your environment is actually going to add that header to traffic from laptops in coffee shops, home offices, and airport lounges. This post covers the mechanism, the four ways to enforce it, and the gaps each one leaves. If you want the background on the identifier itself, start with what a ChatGPT workspace ID is and how to use it for security.

The one-line thesis: tenant restriction for ChatGPT is an HTTP-layer control, so it only works where something can read and write inside the TLS session.

How does OpenAI's tenant restriction for ChatGPT Enterprise work?

Through one custom HTTP header. OpenAI documents the control in its help article on corporate network controls in ChatGPT Enterprise (OpenAI Help Center). The pieces:

• Your network adds the header ChatGPT-Allowed-Workspace-Id to requests bound for ChatGPT.
• The value is your workspace UUID, which admins find in ChatGPT admin settings. Several UUIDs can be listed, comma-separated with no spaces.
• When a user signs in, ChatGPT checks the target workspace against the list. Access to any workspace not on it, including every personal account, is filtered.
• For the iOS, macOS, and Android apps, OpenAI's guidance is to block specific domains at the firewall, proxy, or SASE layer so personal use can't route around the header.

Notice what this requires. The header must be present on the request as ChatGPT receives it. Something in the path has to terminate TLS, add the header, and re-encrypt. That single requirement decides which enforcement options actually work.

Which enforcement methods can block personal ChatGPT?

Four are commonly recommended. Two of them can't do the job. Here's each one, self-contained.

DNS filtering (Cisco Umbrella, DNSFilter)

DNS resolves chatgpt.com identically for the CEO's Enterprise seat and a contractor's free account. It never sees the request, let alone adds a header to it. Cisco's own documentation says that allowing a private ChatGPT instance while blocking others on Umbrella requires the intelligent proxy, SSL decryption, and a root certificate on every device; the DNS layer alone can't do it. If your only control is DNS, your choice is all of ChatGPT or none of it. We covered why that fails in why blocking ChatGPT doesn't work.

Browser extensions and browser-only controls

An extension can watch the ChatGPT web app in one browser. It can't see the ChatGPT desktop app, a second browser, an API call from a script, or an IDE plug-in. Employees find the gap in an afternoon.

Cloud-proxy SWG (Zscaler, Netskope, Cloudflare Gateway)

These do work, because they terminate TLS and can inject the header. Cloudflare documents OpenAI tenant control in its HTTP policies, for example. The cost is architectural: every ChatGPT request now detours from the laptop to the vendor's point of presence and back before it reaches OpenAI. Netskope's own SLA quotes a 50 ms target for decrypted traffic versus under 10 ms for traffic it doesn't decrypt, and tenant control requires decryption. You pay the latency on every token.

On-device SWG (dope.security)

The dope.security agent runs an on-device proxy that performs SSL inspection locally. Cloud Application Control adds the tenant restriction header for your ChatGPT Enterprise workspace on the device itself, so the corporate session proceeds and a personal login is blocked. There's no data-center detour, no shared egress IP, and the control is identical on the office LAN, home Wi-Fi, and a hotel network. The same policy can restrict Claude, Google, and Microsoft 365 to your approved tenants. We shipped this specifically for ChatGPT Enterprise customers; the announcement is in Blocking ChatGPT Personal.

Step by step: enforce ChatGPT Enterprise tenant restrictions with dope.security

• Get the workspace ID. In ChatGPT admin settings, copy your workspace UUID. If you run more than one workspace, collect all of them.
• Deploy the agent. Push dope.endpoint silently via Intune, Jamf, or Kandji. Mac and Windows are supported, and the agent uses under 100 MB of RAM.
• Turn on AI Analytics first. Before you block anything, look at the personal versus enterprise-licensed account breakdown. You'll usually find more personal ChatGPT use than the survey said.
• Enable Cloud Application Control for ChatGPT. Enter the workspace ID. Policy lands on every endpoint in under a minute.
• Add Dopamine DLP. Tenant control decides which account can be used. Dopamine DLP decides what can be pasted or uploaded into it, classifying prompts on the device with zero data retention.
• Handle mobile. Follow OpenAI's guidance and block the personal-app domains for unmanaged mobile devices through your MDM or firewall.

What about the ChatGPT desktop app, API traffic, and other AI tools?

This is where browser-based and DNS-based answers fall apart, and where the endpoint model holds. Because the dope.security proxy sits on the device at the network layer, it inspects traffic from the ChatGPT desktop app and from browsers alike. AI Analytics discovers ChatGPT, Claude, Gemini, Copilot, and shadow AI tools, attributes usage per user, and lets you block an app or enforce enterprise-only access from the same screen. The three-layer model, Shadow IT discovery, SWG policy, then Cloud Application Control, is described in the AI governance stack every CISO should own.

How do the recommended options compare?

• Can it add the OpenAI header at all: DNS filtering and browser extensions cannot; cloud proxies and dope.security can.
• Where enforcement happens: Zscaler, Netskope, and Cloudflare inject the header at a data center after a round trip; dope.security injects it on the laptop with no detour.
• Coverage of desktop apps and IDEs: browser-only tools miss them; dope.security's on-device proxy sees every process on the endpoint.
• Prompt DLP: Zscaler's documentation places prompt DLP in its Data Protection add-on; Netskope's AI Guardrails sit in the higher tiers of its Skylight family; Dopamine DLP is native to dope.security.
• Console: the proxy vendors stack AI controls across multiple products announced in 2026, several with GA dates and pricing not disclosed; dope.security keeps discovery, tenant control, and DLP in one console.
• Analyst standing: Zscaler and Netskope are 2026 Gartner SSE Magic Quadrant Leaders; dope.security's case is architecture and simplicity, not quadrant placement.

When should you choose dope.security?

• You own ChatGPT Enterprise seats and need personal accounts blocked on the same domain, today.
• Your workforce is remote or travels, and a cloud-proxy detour on every AI request isn't acceptable.
• You want the same tenant control for Claude, Google, and Microsoft 365 without a second product.
• You need prompt and upload DLP included, not sold as an add-on.
• Your IT team is small and wants a deployment measured in days.

That last point is real. Greylock Partners went from first proposal to signed contract in 27 days after DNS-only filtering couldn't see the HTTPS traffic that mattered.

The bottom line

OpenAI gives you a clean way to restrict ChatGPT to your workspace. It's an HTTP header, so it has to be enforced inside the encrypted session, on the device or in a proxy. Enforce it on the device and you get the control without the detour. Start a free dope.security trial or book a 20-minute demo.

Frequently Asked Questions

How do I block personal ChatGPT accounts but allow corporate ChatGPT?

Inject OpenAI's ChatGPT-Allowed-Workspace-Id header with your Enterprise workspace ID on all ChatGPT traffic. The header must be added inside decrypted TLS, so use an on-device proxy like dope.security or a cloud proxy; DNS filtering can't do it.

What is the ChatGPT-Allowed-Workspace-Id header?

It's the custom HTTP header OpenAI uses for tenant restrictions in ChatGPT Enterprise. Set its value to one or more workspace UUIDs from your admin settings, and ChatGPT filters sign-ins to any workspace not on the list.

Can Cisco Umbrella block personal ChatGPT while allowing ChatGPT Enterprise?

Not at the DNS layer. Cisco's documentation says this needs the intelligent proxy, SSL decryption, and a root certificate on each device. dope.security performs that inspection on the endpoint with no backhaul.

Does tenant restriction cover the ChatGPT desktop and mobile apps?

The header covers any traffic the enforcing proxy can see. dope.security's on-device proxy covers browsers and the desktop app on managed Mac and Windows devices. For mobile, OpenAI recommends blocking the personal-app domains at the firewall, proxy, or SASE layer.

Does blocking personal ChatGPT also stop sensitive data from reaching ChatGPT Enterprise?

No. Tenant control decides which account is allowed, not what goes into it. Pair it with Dopamine DLP, which classifies prompts and uploads on the device and can block PII, PCI, PHI, and intellectual property before they leave.

Can I allow more than one ChatGPT Enterprise workspace?

Yes. OpenAI's header accepts multiple workspace UUIDs separated by commas with no spaces, so a company with several workspaces can allow all of them while still blocking personal accounts.