How to Block AI Browsers in Your Organisation
AI browsers read the pages your staff open and can click, type and buy inside their signed-in accounts. Gartner told security leaders to block them for now, yet on most networks one download is all it takes.
Four layers that stop an AI browser
Next step
AI Agent Blocklist
AI browser domains, desktop app IDs and agent extension IDs in files your firewall, DNS filter, Intune, Jamf and Chrome policies can read, so the block covers every layer at once.
From $99 a month for one organisation
- Updated every day as new AI browsers appear
- A risk grade on every entry
- Download access set up within one business day
What counts as an AI browser in 2026
The term covers four kinds of software. A rule that names one browser misses the other three, and staff simply move to the next tool. Block the category, not the brand.
Standalone AI browsers
Separate apps with an assistant and an agent built in. They replace Chrome or Edge on the device.
Examples: Perplexity Comet, Dia, Opera Neon, Genspark AI browser, Fellou, BrowserOS, Strawberry, Sigma, Aside.AI inside approved browsers
Gemini in Chrome and Copilot in Edge add a sidebar and, in some editions, agent actions to browsers you already allow.
Controlled with browser policies, not app rules.Agent extensions
Extensions that let an agent read and click on any page turn an approved browser into an agentic one.
Examples: Claude in Chrome, Manus AI Browser Operator, Nanobrowser.Desktop apps with a browser inside
Some AI desktop apps now ship their own browser pane and agent, so the browser is a feature, not a separate install.
Example: the ChatGPT desktop app after ChatGPT Atlas was retired.ChatGPT Atlas is retired, but copies remain. OpenAI announced the end of Atlas on 9 July 2026 and it stopped working from 9 August 2026, with no more security updates. Treat leftover copies on Macs as software that no longer gets security updates. Our ChatGPT Atlas removal guide covers the clean-up.
Why block AI browsers: what goes wrong
An AI browser is two things at once: a sidebar that sees what you see, and an agent that acts as you. Both break assumptions your security controls were built on.
Pages go to the AI service
The sidebar can send the open page, other tabs and browsing history to the vendor's cloud. Anything on screen may leave your control.
Agents act with the user's access
The agent works inside sessions where the user is already signed in: email, CRM, payroll, cloud consoles and banking portals.
Web pages can give orders
Hidden text in a page, a comment or a link can be read as an instruction. This is indirect prompt injection, and it needs no malware.
Wrong actions, real effects
Flawed reasoning can fill in a form wrongly, order the wrong item or book the wrong flight. The bill still arrives.
Phishing gets easier
An agent that follows a link to a fake sign-in page can hand over credentials that the user would have spotted as fake.
Training done by a bot
Staff can ask an agent to click through mandatory security training or routine approvals. The record says done; nobody learned anything.
Researchers showed in 2025 that hidden text in a public comment could steer an AI browser's assistant into reading a user's email address and a one-time code and posting them.
A crafted link carrying instructions could make an AI browser's assistant gather data from the user's connected email and calendar and send it to an outside address.
Text made to look like a web address, pasted into an AI browser's address bar, could be handled as a trusted command from the user.
A page could plant instructions in an AI assistant's memory, where they stayed for later sessions and shaped work done days afterwards.
Instructions hidden after the # in an ordinary web address could reach AI browser assistants, while the website itself never saw them.
These are patterns for the whole category, not findings about one product. Makers describe prompt injection as an open problem for the industry. That is why a policy decision, not a product version, is the real control. More detail is on our AI browser prompt injection page.
Who is exposed, and signs an AI browser is already in use
Risk follows access. The users with the most valuable sessions are often the first to try a tool that promises to do their busywork.
Who is exposed
- Finance: payment portals, supplier records and invoices an agent could change or approve.
- HR and payroll: personal data on screen that a sidebar may send to an AI service.
- Developers and admins: cloud consoles and code hosts where one wrong click changes production.
- Sales and support: CRM records and customer tickets, exported in bulk by a busy agent.
- Executives and assistants: email and calendars, the first target of prompt injection.
- Schools: students using agents for homework. See AI browsers on Chromebooks.
Signs to look for
- A browser in software inventory that IT never packaged.
- Mac process names such as
CometorBrowserOSin running process reports. - Windows package IDs such as
Perplexity.CometorBrowserOS.BrowserOSin install reports. - DNS queries to AI browser and agent domains such as
perplexity.ai,browseros.comormanus.im. - New extensions that can read and change data on all websites.
- Users who changed their default browser after a software update.
- Tickets about pages that "filled themselves in" or orders nobody remembers placing.
How to block AI browsers: the four-layer control map
No single control is enough. A firewall rule misses laptops on home Wi-Fi, an app rule misses extensions, and a browser policy misses a second browser. Used together, the layers close each other's gaps.
Network: firewall and DNS filter
Network blocks stop the download pages and the AI services the browsers call. Load our list as an external feed so it refreshes on its own.
Fortinet FortiGate
- Go to Security Fabric > External Connectors and click Create New.
- Under Threat Feeds choose Domain Name and enter the address of the plain domain list from your download.
- Open your DNS filter profile, find the new feed and set it to block.
- Apply the profile to the outbound policies for user networks and check the logs.
Full walkthrough: block AI agents at the firewall.
Palo Alto Networks firewalls
- Go to Objects > External Dynamic Lists and add a list.
- Choose Domain List for DNS-based blocking in an Anti-Spyware profile, or URL List for a URL Filtering profile.
- Point the source at the matching file from your download. One list holds one type only.
- Reference the list in the profile or a Security policy rule, then commit.
DNS filters and Pi-hole style resolvers take the DNS import or hosts file. See blocking AI with DNS.
Mind the trade-off. Blocking a vendor's main domain can also block its web chat that some teams use. Each entry in our list carries a category and a risk grade, so you can block AI browsers and agents without cutting the chat tools you approved.
Device: stop install and launch
App rules work on any network, including home Wi-Fi and hotel networks. They also stop a browser that is already installed from opening.
Windows with Intune or Group Policy
- In the Intune admin center open Endpoint security > App Control for Business, or use AppLocker under Computer Configuration > Windows Settings > Security Settings > Application Control Policies > AppLocker.
- Build a publisher rule from the signed installer of each AI browser on a test machine.
- Run it in audit mode first, then switch to enforce for a test group.
- Use package IDs such as
Perplexity.CometandBrowserOS.BrowserOSto find and remove copies already installed.
Step by step for Intune: block AI with Intune.
Macs with Jamf Pro
- In Jamf Pro click Computers, then Restricted Software, then New.
- Enter the exact process name, for example
Comet. - Tick Kill process, decide on Delete application, and write a short message that points users to the approved browser.
- Scope the record to all managed Macs and save.
Bundle IDs such as ai.perplexity.comet and at.studio.AsideBrowser help you find Macs that already have the apps. Other Mac MDMs have similar settings; check the current admin console.
Browser policy: extensions and built-in AI
This layer covers the AI that rides inside browsers you already approved. It is where most "block AI in browser" requests end up.
Agent extensions
- For Edge in Intune: Settings catalog, Microsoft Edge\Extensions, then Control which extensions cannot be installed. This sets
ExtensionInstallBlocklist. - For Chrome on Windows and Mac: the same
ExtensionInstallBlocklistpolicy through Intune or Group Policy. - For Chrome users in Google Workspace: Devices > Chrome > Apps & extensions > Users & browsers, pick the organisational unit, add the extension by ID and set it to Block.
- Add IDs such as
fcoeoabgfenejglbffodgkkbkcdhcgfn(Claude),cecngibhkljoiafhjfmcgbmikfogdiko(Manus AI Browser Operator) andimbddededgmcgfhfpcjmijokokekbkal(Nanobrowser).
More on this: block AI browser extensions.
AI built into Chrome and Edge
- Chrome: set
GeminiSettingsto control Gemini in Chrome. - Chrome: set
GeminiActOnWebSettingsto control agent actions on web pages, with allow and block lists of sites if you need exceptions. - Edge: set
EdgeCopilotEnabledto control Copilot in Edge, and check the current Edge policy list for Copilot Mode settings. - Test with a standard user account; policy values and defaults change between versions.
Guides: disable Gemini in Chrome and disable Edge Copilot Mode.
Accounts: agent features in approved AI services
Turn agent features off where you allow AI
- ChatGPT Enterprise: admins can turn off the in-app browser, browser use and computer use in the workspace settings.
- Comet Enterprise, if you chose it: admins decide whether staff may use the Assistant and the Agent, and push Chromium policies through MDM.
- Keep agents away from email. Remove mail connectors from AI tools unless a named team needs them.
- Check each service's data retention settings and turn off what you do not need.
Write it down
- State which browsers are approved and which AI features are on.
- Say how staff can request an exception and who decides.
- Say what happens to unapproved AI browsers found on devices.
- Review the list of exceptions every quarter.
Template and wording: AI agent acceptable use policy.
What to block, what to monitor, what to allow
Most organisations land on a default deny for agentic browsing with a few named exceptions. Use this table as a starting point and adjust it to your risk appetite.
| Item | Default action | Why | Control and file |
|---|---|---|---|
| Standalone AI browsers such as Comet, Dia, Opera Neon, Genspark, Fellou, BrowserOS | Block | An agent with the user's sessions, outside your browser policies | App rules by bundle ID, package ID and process name; domains at the firewall |
| Leftover ChatGPT Atlas copies | Remove | No security updates since August 2026 | Find by bundle ID com.openai.atlas, then uninstall |
| Agent extensions | Block | Read and act on every page in an approved browser | Extension IDs in ExtensionInstallBlocklist |
| AI sidebar in Chrome or Edge | Monitor or turn off | Page content can go to the AI service | Browser AI policies per user group |
| Agent actions in approved browsers | Off by default | Rogue or wrong actions in signed-in sessions | Agent policies, with site allow lists for named teams |
| AI tools your company approved | Allow and log | Business need, contract and data terms in place | Leave their category out of the block, keep logging |
| Agent tools graded High or Severe | Block everywhere | Highest risk entries in the list | Risk-filtered files, so you can block only these if you prefer |
Teams with a strict browser standard often flip the table: allow one managed browser, block every other browser and every extension not on an approved list.
Rollout plan: from audit to full block in two weeks
A staged rollout avoids breaking the AI tools people legitimately use. Each step has one owner and one test.
Inventory
Pull software and extension reports. Search them for the app IDs and process names in the list.
Watch
Load the domain feed in log-only mode. See who reaches AI browser and agent domains, and how often.
Announce
Tell staff what changes, which browser is approved and how to ask for an exception.
Test group
Enforce all four layers for IT staff. Fix anything that breaks before it reaches everyone.
Stay current
Let devices and firewalls pull the updated files. Review exceptions monthly.
One list for every layer
Typing IDs into four consoles by hand does not keep up with a category that grows every month. The AI Agent Blocklist gives each layer the file it reads, and refreshes them every day.
- Plain domain list and hosts file
- CSV and JSON with every column
- Firewall external list format
- DNS filter import files and a PAC file
- Chrome and Edge extension block lists
- macOS and Windows identifiers for Jamf and Intune
Want to see which AI browsers exist today? Browse our list of AI browsers.
Ask us or order a plan
Tell us what you need to block and where your rules live. We reply within one business day.
- Files for firewalls, DNS filters, secure web gateways, MDM and browser policies
- Updated every day as new AI browsers, agents and extensions appear
- Plans for one organisation, for MSPs and for product makers, or a hosted filter we run for you
Blocking AI browsers: questions admins ask
How do I block AI browsers on company laptops?
Use four layers together. Block AI browser and agent domains at the firewall or DNS filter. Stop the apps launching with app control in Intune, Group Policy or Jamf. Block agent extensions by policy, and turn off agent features in AI services you approved.
Can I block AI from my browser?
Yes. On your own computer most browsers let you turn off the AI sidebar and assistant in their settings. On company devices an admin can enforce the same with browser policies, block agent extensions by ID and stop separate AI browsers from installing.
Does Gartner say to block all AI browsers?
Gartner's December 2025 advisory, Cybersecurity Must Block AI Browsers for Now, advises blocking AI browsers with existing network and endpoint controls. Low risk tolerance may mean a longer block. Our Gartner advisory page explains it.
Will blocking a domain stop an AI browser that is already installed?
Partly. Blocking the AI service domains stops the assistant and agent from working on your network, but the browser itself still opens and browses. Pair the domain block with an app rule that stops it from launching, and remove existing installs.
How do I block an AI browser extension?
Add the extension ID to the ExtensionInstallBlocklist policy for Chrome and Edge, through Intune, Group Policy or the Google Admin console. For Firefox use its enterprise policies for extensions. Blocking by ID is exact, so you need the ID of every agent extension you want to stop.
How do I stop AI in Chrome and Edge without blocking the browser?
Use the browser policies. Chrome has GeminiSettings for Gemini in Chrome and GeminiActOnWebSettings for agent actions on web pages. Edge has EdgeCopilotEnabled for Copilot in Edge. Check the current policy lists for new AI settings, because vendors add them often.
Is ChatGPT Atlas still a risk?
OpenAI retired ChatGPT Atlas and it stopped working from 9 August 2026, with no further security updates. Copies can still sit on Macs with saved sessions inside, so remove them. Its browsing and agent features moved into the ChatGPT desktop app and a Chrome extension, which need their own controls.
How do I keep up with new AI browsers?
New AI browsers and agent extensions appeared month after month through 2025 and 2026. A list that updates every day, read automatically by your firewall, DNS filter and device tools, keeps the block current without anyone typing new entries.
Perplexity and Comet are trademarks of Perplexity AI, Inc. OpenAI, ChatGPT and ChatGPT Atlas are trademarks of OpenAI. Google, Chrome, Gemini and Google Workspace are trademarks of Google LLC. Microsoft, Windows, Edge, Copilot and Intune are trademarks of Microsoft Corporation.
Apple, Mac and macOS are trademarks of Apple Inc. Gartner is a trademark of Gartner, Inc. Fortinet and FortiGate are trademarks of Fortinet, Inc. Palo Alto Networks is a trademark of Palo Alto Networks, Inc. Jamf and Jamf Pro are trademarks of Jamf.
Opera and Opera Neon, Dia, Genspark, Fellou, BrowserOS, Strawberry, Sigma, Aside, Manus, Nanobrowser, Claude, Firefox and Pi-hole are trademarks of their respective owners.
Names are used only to identify the products discussed. AI Agent Blocklist is not affiliated with, sponsored or endorsed by these companies.