How to Block ChatGPT on Your Network
Blocking ChatGPT is rarely a single on or off switch. Most organisations pick one of three levels, then apply it on the firewall, the DNS filter and the managed devices so it holds everywhere.
Why organisations block ChatGPT
The main reason is data leakage. Staff paste contracts, code, patient notes or customer lists into a chat box, and that text leaves your control. An AI policy without a technical control behind it tends to be ignored.
Confidential data in prompts
Source code, financial forecasts and personal data typed into a personal account fall outside your contracts, retention rules and audit trail.
Regulatory duties
Health, finance, legal and public sector teams often must prove where client data goes. Unvetted AI accounts make that proof impossible.
Academic integrity
Schools block during exams or for younger year groups, and open access for older students once a classroom policy is in place.
Personal versus company accounts
Many firms pay for a business workspace. They want staff to use it, not a free personal account where admin controls do not apply.
Output risk
Generated text can contain errors, copied material or invented citations. Some teams require review before any AI output reaches a client.
AI policy enforcement
A written acceptable use policy needs a matching rule set. The block, warn or allow decision on the network is how the policy becomes real.
Before you block: write down which teams have an approved AI tool, who signs off exceptions and what users see when they hit the block. A clear block page with a link to the approved tool cuts helpdesk tickets sharply.
ChatGPT domains to block
These are the well-known domains the service runs on, based on the vendor's own network guidance. Domain lists change, so treat this as a starting point and confirm against your own logs.
| Domain | What it does | Block for full block? | Partial block note |
|---|---|---|---|
chatgpt.com and *.chatgpt.com | Main web app, sign-in flow and live conversation connections (including WebSocket hosts such as ws.chatgpt.com) | Yes | Keep open if you use the workspace allow header method |
openai.com and *.openai.com | Company site, account and auth pages, the legacy chat.openai.com address and the developer API | Yes | Developers may need api.openai.com; allow it per group |
oaistatic.com and *.oaistatic.com | Static files such as scripts, styles and images for the app | Yes | Required for the app to load at all |
oaiusercontent.com and *.oaiusercontent.com | File uploads and generated files | Yes | Blocking only this host stops uploads while chat still works |
| Voice traffic (UDP 3478) | Real-time voice mode uses its own ports and published IP ranges | Handled by the domain block in most setups | Block UDP 3478 to the published ranges if you allow text but not voice |
The file upload host is the most useful single lever for a partial block. Blocking oaiusercontent.com stops users from dropping spreadsheets and PDFs into the chat, which is where most bulk data leaves the company.
Not sure how a host is categorized today? Run it through the website category lookup to see its category in our database before you write rules around it.
Partial block options that keep work moving
A full block is simple, but it pushes people to phones and home laptops. These partial patterns let you keep the useful parts and close the risky ones.
| Pattern | How it works | Needs | Good fit |
|---|---|---|---|
| Workspace only | Your proxy adds a header to requests for chatgpt.com that lists your approved ChatGPT Enterprise workspace IDs. ChatGPT filters out personal and other workspaces itself. | TLS inspection on the proxy or firewall; an Enterprise workspace. Check the vendor's current admin documentation for the header name. | Companies that pay for a business workspace |
| Group-based access | Allow AI / LLM Tools for named groups (for example engineering), block it for everyone else. | User identity on the firewall or filter (directory sync or agent) | Firms piloting AI with one team first |
| No uploads | Block the upload host and file transfer to AI domains while text chat stays open. | Domain rules; optional DLP on the proxy | Teams where documents are the main risk |
| Warn page | Show a policy reminder with a continue button; log each click-through. | Filter with a "warn" or "continue" action | Organisations still writing their AI policy |
| Isolation | Open ChatGPT in remote browser isolation with paste and upload disabled. | A browser isolation service | High-security teams that still want reading access |
| Time windows | Block during exams or class time, open after hours. | Scheduled rules | Schools and training centres |
Watch the side doors. A partial block of ChatGPT means little if the same staff can use any of thousands of other chat assistants. Pair every ChatGPT rule with a category rule for unapproved AI tools.
Methods: where to put the block
Use more than one layer. The network stops unmanaged devices on your Wi-Fi; endpoint and browser policies follow managed laptops home; MDM handles the phone app.
Router or small office gateway
- Open the router admin page and find parental controls, access control or website blocking.
- Add chatgpt.com, openai.com, oaistatic.com and oaiusercontent.com as blocked domains, not full URLs.
- If the router can set its own DNS server, point it at a filtering resolver, which is more reliable than keyword blocks.
- Save, reboot clients and test.
Keyword filters on consumer routers often fail on encrypted traffic. Treat this as a home or tiny office method only.
Next-generation firewall
- Create a URL or web filter profile with a custom list holding the ChatGPT domains with wildcards.
- Also set your firewall's AI category to block or warn so look-alike tools are caught.
- If the firewall has an application signature for ChatGPT, add it to an application control rule as a second check.
- Enable TLS inspection if you need upload blocking or the workspace header; otherwise SNI-based blocking still works for the whole domain.
- Attach the profile to outbound policies for staff and guest networks.
DNS filter
- Add the domains to the deny list with subdomains included.
- Turn on the AI tools category for the same policy.
- Force all clients to your resolver: block outbound port 53 and 853 to other servers and block known DNS over HTTPS endpoints.
- Return a block page address so users see why it failed.
DNS cannot see paths, so it cannot do "chat yes, uploads no" unless uploads sit on their own host, which they do here. See DNS filtering vs web filtering.
Microsoft Intune
- Use Intune to deploy your endpoint web filter with the ChatGPT domains on its block list, so the rule follows the laptop off site.
- Make sure the policy covers every installed browser, not only Edge.
- For Edge, add the domains to the URL block list policy in the Intune settings catalog.
- Block or remove the desktop app with app control or an uninstall assignment.
Google Admin console
- Go to Devices, Chrome, Settings, Users and browsers, then URL blocking.
- Add chatgpt.com, openai.com, oaistatic.com and oaiusercontent.com to the blocked list for the chosen organisational unit.
- Block ChatGPT related extensions in the Chrome Web Store settings for the same unit.
- Remember this only covers managed Chrome sessions; pair it with a network rule.
MDM for phones and tablets
- On supervised iOS devices, hide or block the ChatGPT app and add a web content filter payload with the domains.
- On Android Enterprise, allow only approved apps from managed Google Play so the app cannot be installed in the work profile.
- Use a per-app VPN or a device DNS profile so mobile data also passes through your filter.
Mobile app vs browser: two different problems
The browser version lives on your network rules. The apps carry their own sign-in and keep working off-site, so they need device controls.
Browser
- Stopped by domain blocks on firewall, DNS or browser policy
- Upload host can be blocked on its own
- Shows your block page, so users know why
- Extensions that embed chat need a separate extension policy
- Covered at home only if the endpoint agent or browser policy travels with the device
Desktop and mobile apps
- Use the same domains, so network blocks work while on your Wi-Fi
- On mobile data the network rule does nothing
- Fail with a vague connection error, not your block page
- Need MDM: app block lists, managed app stores or uninstall assignments
- Desktop apps can be blocked with application control policies
Bring your own device staff are the hardest group. On personal phones you can only control traffic on your Wi-Fi, so keep the guest and staff networks on the same AI policy.
VPN, proxy and other bypass routes to close
Users who want ChatGPT and find it blocked will try a handful of predictable routes. Each one has a matching control. Never rely on one layer alone.
Personal VPN apps
Encrypted tunnels hide the destination from DNS and URL filters.
Web proxies and mirror sites
Small sites resell access to chat models under a new domain every few weeks.
DNS over HTTPS
Browsers can resolve names over encrypted DNS and skip your resolver.
Private relay services
Some phone and browser privacy relays hide traffic from network filters.
Mobile hotspot
A laptop on a phone hotspot leaves your network entirely.
Embedded AI in other apps
Plugins, extensions and third-party apps that call the API from their own servers.
How to test the ChatGPT block
Test from the user's point of view, on each device type and network segment. A rule that works on the admin laptop often fails on guest Wi-Fi or a phone.
- Open chatgpt.com in a fresh private window on a managed laptop. You should see your block page or warn page, not a timeout.
- Try the legacy chat.openai.com address and the openai.com sign-in page. Both should follow the same rule.
- Run a DNS query for chatgpt.com from a client. With a DNS block, the answer should be your block address or no record.
- For a partial block, sign in with a personal account. With the workspace header in place, only the approved workspace should be usable.
- Try to upload a file. With the upload host blocked, chat works but the upload fails.
- Open the phone app on staff Wi-Fi, then on mobile data, on a managed and an unmanaged phone.
- Switch on a consumer VPN and repeat. If it connects, your VPN category rule or app restriction is missing.
- Check the logs. Each attempt should appear with user, device and the rule that matched.
- Visit two or three smaller chat assistants. If they load, your category rule is not doing its job.
Why category-based blocking beats a list of ChatGPT domains
Blocking one brand leaves the rest of the market open. New chat tools, wrappers and resellers appear every week, many on fresh domains, so a static list goes stale fast.
A new chat tool launches on a fresh domain.
It is categorized into AI / LLM Tools, Text & Language, within the daily delta.
It is checked again after a week and after a month, as content grows.
Your filter picks up the update and the same policy applies with no manual list edits.
Vendors and MSPs use the AI / LLM Tools category to build exactly the partial policies on this page: allow approved assistants, warn on document tools, block companions and proxies.
The data comes as files that run inside your own product, with exports such as DNS RPZ, firewall external dynamic lists and hosts files described in our blocklist formats guide.
If you run filtering for many clients, the MSP web filtering page covers per-tenant policies, and corporate web filtering covers the enterprise view.
A sample AI policy matrix you can adapt
Most teams land on a small table like this. Write yours down before touching the firewall, then build one rule per row.
| Group | ChatGPT | Other AI chat tools | File uploads to AI | Logging |
|---|---|---|---|---|
| All staff | Company workspace only | Warn | Blocked | Yes |
| Engineering pilot | Company workspace and API | Approved coding tools allowed | Allowed to approved tools | Yes |
| Finance and HR | Blocked | Blocked | Blocked | Yes |
| Guest Wi-Fi | Allowed | Allowed except proxies | Allowed | Minimal |
| Students (school) | Blocked or class-time block | Approved tutors only | Blocked | Yes |
Frequently asked questions
Can I block ChatGPT but still allow our company ChatGPT workspace?
Yes, on networks with TLS inspection. ChatGPT Enterprise supports a workspace allow header that your proxy or firewall adds to requests to chatgpt.com. Personal and unapproved workspaces are then filtered out by ChatGPT itself. Check the current admin documentation for the exact header name.
Is blocking chatgpt.com enough?
No. The service also uses openai.com subdomains plus static and upload hosts such as oaistatic.com and oaiusercontent.com. Users can also switch to another chat assistant, which is why a category rule matters.
Does blocking ChatGPT on the network stop the mobile app?
Only while the phone uses your network. On mobile data the app connects directly, so managed devices also need an MDM rule that removes or restricts the app.
Why does ChatGPT still load after I added a DNS block?
Usually the browser uses DNS over HTTPS, the device cached the old answer, or the user is on a VPN or private relay. Disable DoH by policy, flush caches and block the VPN and proxy category.
Will blocking openai.com break other software?
It can. Some business apps call the API from the user's device. Check your logs for api.openai.com traffic from approved software before a full block, and allow it per group if needed.
Which category in your database covers ChatGPT?
AI / LLM Tools, in the Text & Language subcategory. The set covers 16,000+ AI-tool domains in 18 subcategories, updated daily. Check any host with the domain lookup tool.
Is it legal to block ChatGPT at work?
Employers can generally set rules for company networks and devices. Monitoring and logging rules differ by country, so check local employment and privacy law before you log individual use.
Keep your AI rules current without editing lists
License a daily-updated category database with 16,000+ AI-tool domains in 18 subcategories, inside your own firewall, DNS filter or product.
ChatGPT and OpenAI are trademarks of OpenAI. Microsoft, Intune and Edge are trademarks of Microsoft Corporation. Google, Chrome and Chromebook are trademarks of Google LLC. iCloud is a trademark of Apple Inc.
These names are used here only to identify the products discussed. Web Filtering Database is not affiliated with, sponsored or endorsed by any of these companies.