Block AI agents

How to block AI agents on your network and devices

AI agents now act for your staff: they browse, send mail, move files and run code with the user's own access. Most of them were never approved by IT.

There is no single switch. You stop them in the five places they live, with one list behind every rule.

Network

Agent services and their domains

Browser

Agent extensions and sidebars

Devices

AI browsers and agent apps

Developer tools

Coding agents, CLIs, MCP servers

Built-in agents

Assistants inside apps you already use

Next step

AI Agent Blocklist

One daily-updated list of AI agent domains, agent extensions, AI browsers, agent apps and coding agents, with a risk grade on every entry, in files for your firewall, DNS filter, MDM and browser policies.

From $99 a month for one organisation

Why "block AI" in one place is not enough

An AI chatbot answers questions. An AI agent does things: it opens sites, fills forms, sends messages and runs commands. That is why a single URL category or a single app rule leaves gaps.

2,600+

AI agent domains in our list, from AI browsers to MCP servers

5,300+

agent identifiers on devices: extension IDs, bundle IDs, package IDs, process names

5

places an agent can live, each with its own enforcement tool

Daily

updates as new agents, browsers and extensions appear

A domain block misses local agents

An agent that runs on a local model, or that reuses a session already open in the browser, keeps working when its website is blocked. You also need the app and extension identifiers.

An app block misses the web

Many agents need nothing installed. They run in a browser tab or a cloud workspace and reach your systems through the user's signed-in accounts. You also need the domains.

Generic AI categories are too blunt

Blocking every AI site also blocks the approved assistant your company pays for. Agents need their own list, so you can block agents and keep the chat tool.

Agents arrive faster than reviews

New AI browsers, extensions and MCP servers appear every week. A list you wrote by hand last quarter is already out of date. Ours updates every day.

The five places AI agents live, and how to block each

Read across each row: what lives there, where you enforce, and what our list gives you for that tool. The linked guides go deeper on each layer.

1. Network

What lives here

The services behind agents: autonomous agent platforms, deep-research agents, AI sales and meeting agents, agent builders, hosted MCP servers.

Where you enforce

Firewall, DNS filter or secure web gateway. Our guides cover firewalls and DNS filters.

What our list gives you

Agent domains as a plain list, firewall external list, hosts file, PAC file and DNS import files. Examples: manus.im, browser-use.com.

2. Browser

What lives here

Extensions that read every page and act in it: browser agents, AI sidebars with agent modes, automation helpers.

Where you enforce

The extension block list policy in Chrome and Edge through Group Policy, Intune or the Google Admin console. See blocking AI browser extensions.

What our list gives you

Extension IDs ready for the policy. Examples: Claude fcoeoabgfenejglbffodgkkbkcdhcgfn, Nanobrowser imbddededgmcgfhfpcjmijokokekbkal.

3. Devices

What lives here

AI browsers and desktop agent apps that users install themselves, often without admin rights.

Where you enforce

Your MDM and app control: Intune, Jamf, Kandji, Mosyle, AppLocker. See blocking AI with Intune and blocking AI browsers.

What our list gives you

macOS bundle IDs, Windows package IDs and process names. Examples: ai.perplexity.comet, im.manus.desktop, Perplexity.Comet.

4. Developer tools

What lives here

Coding agents with shell access, command-line agents installed from package managers, MCP servers that give agents new powers.

Where you enforce

Endpoint app control, package manager rules, editor policies and MCP allow and deny lists. See blocking MCP servers.

What our list gives you

Package names and process names of command-line agents. Examples: @anthropic-ai/claude-code, @openai/codex, process cursor-agent.

5. Built-in agents

What lives here

Agent features inside products you already run: assistants in the operating system, office suite, browser, chat apps and meeting tools.

Where you enforce

Each product's own admin settings. For Microsoft, see turning off Copilot actions and agents.

What our list gives you

The web services these features fall back to, so a disabled feature is not replaced by its website. The admin setting stays your first control here.

Block every agent, or only the high-risk ones

Not every company wants a full ban. Every entry in our list carries one of five risk grades, so "block high risk AI agents" is a filter on a column, not a research project.

Low

The lowest concern in our grading. Usually monitored, not blocked.

Moderate

A typical level of concern. Block if not approved.

Elevated

Higher concern. Block by default in most companies.

High

Serious concern. Block on every device and network.

Severe

The highest concern. Block and look for existing installs.

Strict

Block all agents

Import every entry. Approved tools go on an allow list that overrides the block.

Fits regulated teams, schools and anyone following Gartner's December 2025 advice on AI browsers.

Tiered

Block from Elevated up

Block Elevated, High and Severe. Log Low and Moderate, and review the top hits each month.

Fits most companies that are still writing their agent policy.

Allow first

Allow only approved agents

Deny all agents by default and allow a short named list. Our list tells you what the default must cover.

Compare both models in allowlist or blocklist for AI agents.

A four-week rollout that does not break work

Blocking agents on day one without warning sends people to their phones. Most teams get better results with a short, staged rollout.

  1. Week 1: See

    Load the list into your DNS filter and firewall in log-only mode. Run the device identifiers against your software inventory. You now know which agents are in use and by whom. Our guide to finding shadow AI agents helps read the results.

  2. Week 2: Decide

    Pick a strategy: strict, tiered or allow first. Name the agents you approve and the accounts they may use. Tell staff what changes and why.

  3. Week 3: Block

    Switch the network rules to block. Push the extension block list and the app rules. Turn off built-in agent features you do not use in each product's admin settings.

  4. Week 4: Hold

    Check for reinstalls and new agents. Keep the daily list update running so new tools are blocked the day they appear. Review exceptions every quarter.

Quick answers for specific agents

These are the questions admins search for most. Settings in these products change often, so check the current admin console before you roll anything out.

How to disable the AI agent in VS Code

A user can switch off agent mode with the chat.agent.enabled setting. On managed machines, set the matching policy so it stays off.

VS Code reads policies from Group Policy on Windows (stored under Software\Policies\Microsoft\VSCode), from a configuration profile on macOS, and from /etc/vscode/policy.json on Linux from version 1.106.

PolicySetting it locksUse it to
ChatAgentModechat.agent.enabledStop agent mode from being turned on in chat
ChatMCPchat.mcp.accessControl access to installed MCP servers
ChatToolsAutoApprovechat.tools.global.autoApproveKeep manual approval of tool runs in place
AllowedExtensionsextensions.allowedAllow only approved extensions, which stops third-party agent extensions

Newer VS Code versions also have policies for MCP allow and deny lists and for third-party agent sessions. Our page on coding agent security risks covers what else to lock on developer machines.

How to disable the AI agent in Replit

Replit's agent is part of the workspace itself. We found no single published switch that turns it off for a whole organisation, so check your plan's current admin settings.

  • If Replit is not approved, block replit.com at DNS or the firewall.
  • If it is approved, ask users to have the agent plan first and wait for approval before it builds.
  • Use its checkpoints to roll back changes the agent made.

How to stop the AI agent on WhatsApp

We found no published setting that removes Meta AI from WhatsApp. What you can control:

  • Turn on Advanced Chat Privacy in sensitive chats and groups (tap the chat name). It stops others using that chat's messages for AI features.
  • On company phones, manage WhatsApp itself with your MDM.
  • Block meta.ai on the network if it is not approved.

On WhatsApp Business, an AI agent that answers your customers is switched off where it was set up: in the app's business tools or your messaging provider's console.

How to block AI agents from your website

This is the opposite problem: outside agents and bots visiting your site, not your staff's agents leaving the network.

  • Use robots.txt for the AI bots that honour it. It is a request, not a lock.
  • Add user agent and rate rules on your web server for the rest.
  • Expect some AI browsers acting for a person to look like a normal visitor.

Our list is built for outbound control, so it is not a tool for this job.

How to stop AI agents going rogue or acting on hallucinations

An agent that misreads a page or invents a fact can still click "send". Limit what a wrong answer can do:

  • Give approved agents the fewest permissions that work.
  • Require a human to confirm sends, payments, deletions and posts.
  • Keep agents out of mail, finance, HR and admin consoles.
  • Log agent actions and review them.
  • Block every agent you have not reviewed. Unreviewed agents cause most surprises.

Five mistakes that let agents back in

We see the same gaps again and again when a company first tries to stop AI agents. Each one has a simple fix.

Blocking only the main website

Agents talk to API hosts, sign-in hosts and update hosts, not just the homepage. Block the full set of agent domains, not the one name everyone knows.

Forgetting per-user installs

Many AI browsers install into the user's own profile with no admin rights. Inventory that only looks at system folders will not see them.

Policing one browser

An extension block in Chrome does nothing in Edge, Brave or a second Chromium browser. Apply the extension list to every managed browser and block the unmanaged ones as apps.

Leaving sessions alive

Blocking an agent today does not end the tokens it collected yesterday. Revoke its sign-ins to mail, storage and SaaS tools when you remove it.

Writing the list once

A hand-made list is out of date within weeks, because new agents, extensions and MCP servers appear all the time. Point your firewall, DNS filter and MDM at a list that updates every day, and keep exceptions in a separate allow list you own.

What to block, what to monitor, what to allow

A starting position most IT and security teams can defend to the board, to staff and to auditors. Adjust it to your own risk appetite.

Block

  • AI browsers on managed devices
  • Browser agent extensions
  • Autonomous agent platforms not approved
  • Command-line coding agents outside developer groups
  • Unapproved MCP servers and hosted agent tools
  • Everything graded High or Severe

Monitor

  • Low and Moderate agent domains, in log mode
  • New installs of agent apps in user profiles
  • Agent use on developer machines
  • AI meeting bots joining calls
  • Daily list changes for new agents

Allow

  • The assistant your company licensed, through company sign-in
  • Named coding agents for named teams, with editor policies set
  • Agents that confirm before acting and keep logs
  • Each exception with an owner and a review date

Everything above, in one download

The AI Agent Blocklist is built from our AI tools data of 12,000+ AI tool domains.

It covers AI browsers and browser agents, general autonomous agents, personal AI assistants, agent builders, multi-agent orchestration, MCP servers, coding agents, deep-research agents, meeting and voice agents, AI sales agents and AI RPA.

Every entry has the tool name, its category, the type of identifier and a risk grade. After an order we set up your download access within one business day.

From $99 a month for one organisation

Domain filesPlain list, hosts file and PAC file for proxies and endpoints
Firewall feedExternal list format for FortiGate, Palo Alto and other firewalls that read a URL list
DNS filter importsFiles for DNS filters, Pi-hole and AdGuard Home
Browser policiesChrome and Edge extension block lists for Group Policy, Intune and the Google Admin console
Device identifiersmacOS bundle IDs and Windows package IDs for Jamf and Intune, plus process names
Full dataCSV and JSON with every column, including category and risk grade

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

We use your details only to answer you. No newsletters.

Questions about blocking AI agents

How do I block AI agents in my company?

Block them where they live: agent domains at the firewall or DNS filter, agent extensions in your browser policy, AI browsers and agent apps with your MDM, and coding agents and MCP servers on developer machines.

Switch off built-in assistants with each product's own settings. One list that covers all five places keeps the rules in step.

Can I block only high-risk AI agents?

Yes. Every entry in our list carries a risk grade from Low to Severe. You can import only the High and Severe entries, or block everything from Elevated up and watch the rest, then tighten later.

How do I disable the AI agent in VS Code?

Users can turn off the chat.agent.enabled setting. For managed machines, set the ChatAgentMode policy through Group Policy on Windows, a configuration profile on macOS or the policy file on Linux, so users cannot turn it back on.

Related policies control MCP servers, tool auto-approval and allowed extensions.

How do I disable the AI agent in Replit?

Replit's agent is part of the workspace, and we found no single published switch that turns it off for a whole organisation. If Replit is not approved, block its domain.

If it is approved, ask users to have the agent plan first and wait for approval, and use checkpoints to roll back changes. Check your plan's current admin settings.

How do I stop the AI agent on WhatsApp?

We found no published setting that removes Meta AI from WhatsApp. In sensitive chats and groups, turn on Advanced Chat Privacy, which stops others using the chat's messages for AI features.

On company phones, manage WhatsApp itself with your MDM, and block the Meta AI website on the network if it is not approved.

How do I block AI agents from my website?

That is the opposite problem: outside agents and bots visiting your site. Use robots.txt for the bots that honour it and user agent or rate rules on your web server for the rest. Our list is for the other direction: stopping agents your own staff run from reaching company systems.

How do I stop AI agents going rogue?

Give approved agents the fewest permissions that work, require a human to confirm sends, payments and deletions, keep agents out of mail, finance and admin tools, log what they do, and block every agent you have not approved. Most rogue actions come from agents nobody reviewed.

VS Code, Intune, Edge, AppLocker and Windows are trademarks of Microsoft Corporation. Chrome is a trademark of Google LLC. Claude and Claude Code are trademarks of Anthropic, PBC. Codex is a trademark of OpenAI.

Perplexity and Comet are trademarks of Perplexity AI, Inc. Manus, Nanobrowser, Cursor, Replit, WhatsApp and Meta AI are trademarks of their respective owners.

Jamf, Kandji, Mosyle, FortiGate, Palo Alto Networks, Pi-hole and AdGuard Home 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.