AI Agent Blocklist: domains, extensions and apps of AI agents in one feed
AI agents now click, type, buy and run commands on company devices. Most arrived without a review. One daily feed lets your firewall, DNS filter and device management stop them before they reach your data.
The AI Agent Blocklist: every agent domain, extension, app and coding CLI in files your controls already read
Domains for the firewall and DNS filter, extension IDs for Chrome and Edge policies, app and package identifiers for Intune and Jamf. Each entry has a category and a risk grade.
Why blocking AI agents is harder than blocking a website
A chatbot answers questions. An agent acts. It signs in with the user's sessions, reads mail and files, fills forms and runs commands. It also arrives through six different doors, and a web filter covers only one of them.
AI browsers
Whole browsers built around an assistant that reads every page and can act in logged-in tabs. Installed per user, often without admin rights.
Agent extensions
Extensions that add an agent to Chrome or Edge. They can see and change every page the user opens, including mail, CRM and banking.
Desktop agents
Apps that control the mouse, keyboard and screen, or run tasks in a cloud browser with the user's accounts connected.
Coding agents and CLIs
Command line agents installed from package managers. They read source code and secrets and run shell commands on the developer's machine.
MCP servers
Small connectors that give an agent access to a database, a mailbox, a file share or a cloud console. Anyone can start one locally.
Agent platforms
Web services where staff build their own agents, connect company accounts and let them run on a schedule, outside any review.
What goes wrong when agents run unmanaged
| What happens | How it happens | Who is exposed |
|---|---|---|
| Data leaves through the agent | The agent reads open tabs, mail or files and sends them to its AI service to work on them. | Anyone who handles customer, patient, student or financial data |
| The agent follows someone else's orders | Hidden text on a web page or in a document tells the agent what to do. Researchers showed this many times in 2025. | Every user of an AI browser or agent extension |
| Wrong actions in real accounts | The agent misreads a page and submits the wrong form, order or payment in a logged-in session. | Finance, procurement, HR and sales teams |
| Destructive commands | A coding agent with shell access deletes files or changes systems after a bad instruction. | Developers, DevOps, IT admins |
| Unapproved connections to company systems | An MCP server or agent platform is given tokens to mail, storage or cloud consoles. | Owners of every connected system |
| Training and audits completed by a bot | Staff let an agent click through mandatory courses or forms on their behalf. | Compliance, HR, schools |
Gartner's advice: in December 2025 Gartner published "Cybersecurity Must Block AI Browsers for Now" and advised blocking AI browser installs with existing network and endpoint controls. Read our summary on what the Gartner advice means for IT teams.
Signs agents are already in use
- DNS or proxy logs show traffic to agent platforms nobody approved
- Browser reports list new extensions that request access to all sites
- Software inventory shows browsers your team never packaged
- Developers' machines run command line tools installed from npm or PyPI
- Calendar invites include AI notetakers or voice agents
- Service accounts or tokens appear for tools that have no owner
What the AI Agent Blocklist contains
The list is built from our AI tools data, which covers more than 12,000 AI tool domains. From it we take every tool that acts on its own or helps others build agents: 2,600+ domains and 5,300+ identifiers on devices.
AI browsers and browser agents
Agentic browsers and the services that drive a browser for a user.
Example:browser-use.comGeneral autonomous agents
Agents that take a goal and work through many steps on their own.
Example:manus.imPersonal AI assistants
Assistants that act in the user's mail, calendar and accounts.
Example:lindy.aiAgent building platforms and frameworks
Services and toolkits for building and hosting agents.
Example:langchain.comMulti-agent orchestration
Tools that run teams of agents that hand work to each other.
Example:crewai.comMCP servers and integrations
Connector hubs and servers that link agents to data and systems.
Example:glama.aiAutonomous coding agents
Agents that write, change and run code with little supervision.
Example:devin.aiDeep research agents
Agents that browse, read and compile reports over many sources.
Example:elicit.orgAI meeting and voice agents
Notetakers that join calls and voice agents that answer phones.
Example:otter.aiAI sales agents
Agents that prospect, write and send outreach on their own.
Example:aisdr.comAI RPA
Automation tools that click through web apps for the user.
Example:bardeen.aiMixed tools
A tool can sit in more than one category. Each entry lists all of them.
See the list of AI browsersIdentifier types on devices
Domains stop traffic. Identifiers stop installs and launches, and they also work when a laptop is off your network.
| Identifier type | What it identifies | Where you use it |
|---|---|---|
| Chrome extension ID | An agent extension in Chrome and other Chromium browsers | Chrome policy, Group Policy, Intune, Google Admin console |
| Edge add-on ID | An agent extension in Microsoft Edge | Edge policy through Group Policy or Intune |
| Firefox add-on ID | An agent extension in Firefox | Firefox enterprise policies |
| macOS bundle ID | An AI browser or agent app on a Mac | Jamf Pro, Kandji, Mosyle, Intune scripts |
| Windows package ID and Microsoft Store ID | An AI browser or agent app on Windows | Intune inventory, detection and removal scripts |
| Process name | The running program on macOS, Windows or the command line | Jamf Restricted Software, endpoint rules, hunting queries |
| Package name (npm, PyPI) | A command line coding agent or MCP server package | Package proxy rules, developer machine checks |
| IDE extension ID | Coding agents inside VS Code and JetBrains IDEs | IDE extension policies and allowlists |
The columns you get, with real rows
The CSV and JSON files carry every column. Below are six real entries from the current data. The full files hold thousands of rows like these.
| Tool | Category | Identifier type | Value | Risk grade |
|---|---|---|---|---|
| Manus AI Browser Operator | Agents and automation: general autonomous agents | Chrome extension ID | cecngibhkljoiafhjfmcgbmikfogdiko | Graded |
| Comet (Perplexity) | Search: AI search and answer engines | macOS bundle ID | ai.perplexity.comet | Graded |
| BrowserOS | Agents and automation: browser agents | Windows package ID | BrowserOS.BrowserOS | Graded |
| Gemini CLI | Code and development: autonomous coding agents | npm package | @google/gemini-cli | Graded |
| goose | Agents and automation: general autonomous agents | Command line process | goose | Graded |
| Browser Use | Agents and automation: browser agents | Domain | browser-use.com | Graded |
Grades belong to the tool, not to your use of it. A Moderate tool can still be wrong for a team that handles patient records, so the grade is a starting point for your rules.
Block high risk AI agents first: the risk grade
Every entry carries one of five grades. Use them to block in stages: the riskiest tools on day one, the rest after you have talked to the teams who use them.
Block everywhere, on and off the network. No exceptions without a written decision.
Block by default. Allow only for a named group with a business reason.
Block or require approval. Review who uses it before you decide.
Allow only approved tools. Block the rest of the category.
Monitor. Block if your policy says no agents at all.
Block all agents
Load the full domain list and the full identifier files. This fits schools, regulated firms and anyone following the Gartner advice to the letter.
Block by grade and category
Filter the CSV by grade or category and load only those rows. Keep approved tools on an allowlist. Our guide on allowlists versus blocklists helps you choose.
Formats and where each one is used
You download the same data in the shape each control expects. No conversion, no manual typing of IDs.
| Format | What is in it | Where it goes |
|---|---|---|
| Plain domain list | One agent domain per line | DNS filters, Pi-hole, AdGuard Home, any tool that reads a text list |
| Firewall external list | Domains or URLs in the format firewalls fetch | FortiGate threat feeds, Palo Alto Networks external dynamic lists, other firewalls that read a URL list |
| DNS filter import files | Domains ready for a custom destination list | Cisco Umbrella and other DNS filters with list import |
| Hosts file | Domains pointed to a null address | Single machines, test labs, kiosks |
| PAC file | Proxy rules that block agent domains | Browsers and proxies that use automatic proxy configuration |
| Chrome and Edge extension block lists | Extension IDs as policy values | Group Policy, Intune, Google Admin console |
| macOS and Windows identifiers | Bundle IDs, package IDs, process names | Jamf Pro, Intune, other device management |
| CSV and JSON | Every entry with every column | SIEM lookups, inventory reports, your own scripts |
How the blocklist plugs into the tools you already run
Three short recipes. Menu names change between versions, so check the current admin console when a label looks different.
Firewall
- FortiGate: go to Security Fabric > External Connectors > Create New and pick Threat Feeds > Domain Name.
- Point it at your list and select the feed in a DNS filter profile.
- Palo Alto Networks: add a Domain List or URL List under Objects > External Dynamic Lists and use it in your security profiles.
DNS filter
- Create a custom block list or destination list in your DNS filter.
- Import the DNS filter file or add the plain domain list.
- Attach it to the policy for staff, guests or students.
Devices and browsers
- Intune: Settings catalog, Microsoft Edge\Extensions, "Control which extensions cannot be installed", paste the IDs.
- Google Admin console: Devices > Chrome > Apps & extensions > Users & browsers, block by ID.
- Jamf Pro: Computers > Restricted Software, add the process names with Kill process.
Agent extensions get their own walkthroughs for the Google Admin console and Jamf on Macs. Coding agents are covered in blocking Claude Code, and MCP in blocking MCP servers.
Updated every day
New AI browsers, agents, extensions and MCP servers appear every week. A list you build once is out of date within days.
A new agent, extension or CLI is released or renamed.
Every entry is checked before it enters the list, with its category and grade.
The next daily files include it in every format.
Firewalls and DNS filters that read a list pick it up on their own schedule.
Device policies need a push. Extension and app identifiers live in policies. Re-import the new files on a schedule that suits your change process, for example once a week.
What to block, what to monitor, what to allow
A blocklist works well next to a short allowlist of tools you approved. Here is the split most organisations start with.
Block
AI browsers, agent extensions with access to all sites, unapproved coding CLIs, public MCP hubs and every Severe or High entry.
Monitor
Moderate and Low entries in categories you have not ruled on yet. Use the CSV as a lookup in your logs to see who uses what.
Allow
The assistants and coding tools you licensed, with their enterprise settings. Remove their rows from your copy of the block files.
More on the policy side in our guide to blocking AI agents.
Plans and prices
Three plans. All include every format, daily updates and every category. After an order we set up your download access within one business day.
One organisation
- One company, school or district
- All formats and identifier files
- Internal use
MSPs and multi-site
- Separate files per client or site
- All formats and identifier files
- Use for the clients you manage
Product licence
- Embed the data in a security, browser, filter or MDM product
- Terms agreed per product
Block the agents, not the people who need AI
Load the files, keep your approved tools on an allowlist and let the daily updates catch what arrives next.
- 2,600+ AI agent domains with categories and risk grades
- 5,300+ extension, app, package and process identifiers
- Firewall, DNS, hosts, PAC, CSV, JSON and policy files
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
Questions about the AI Agent Blocklist
How do I block AI agents on company devices?
Block them in three places at once: agent domains at the firewall or DNS filter, agent extensions with a browser extension blocklist, and agent apps and command line tools with your device management. The AI Agent Blocklist gives you a file for each of those places.
Can I block only high-risk AI agents?
Yes. Every entry carries a risk grade from Low to Severe. You can load only the High and Severe entries, or block everything from Elevated upwards and review the rest.
Does the list work as an AI DNS blocklist?
Yes. The plain domain list and the DNS filter import files load into DNS filters such as Cisco Umbrella, Pi-hole and AdGuard Home as a custom or external list.
Can I block all AI sites, not only agents?
The AI Agent Blocklist covers AI agents and the tools around them. It is built from our wider AI tools data of more than 12,000 AI tool domains. If you need a broader AI block, tell us in the form and we will explain the options.
Is this a list for blocking AI crawlers from my website?
No. This list stops your own staff and devices from reaching AI agents. Keeping AI crawlers away from a public website is a different job that you handle on the web server and in robots.txt.
Can I block Goose and other open source agents?
Yes. Open source agents are listed with their domains and with the identifiers they leave on devices, for example the Windows package ID AgenticAIFoundation.gooseCLI and the command line process goose.
How often is the AI Agent Blocklist updated?
Every day. New AI browsers, agents, extensions and MCP servers are added as they appear, and every entry is checked before it enters the list.
Can we publish the list or put it on GitHub?
No. The licence covers internal use in your organisation, or use inside your product with a product licence. Resale or publication of the raw lists is not allowed. See the pricing page for licence terms.
Perplexity and Comet are trademarks of Perplexity AI, Inc. Manus is a trademark of its owner. Gemini, Chrome and Google are trademarks of Google LLC. Microsoft, Edge, Intune and Windows are trademarks of Microsoft Corporation. Firefox is a trademark of the Mozilla Foundation. macOS is a trademark of Apple Inc. FortiGate is a trademark of Fortinet, Inc. Palo Alto Networks is a trademark of Palo Alto Networks, Inc. Cisco Umbrella is a trademark of Cisco Systems, Inc. Jamf, Kandji, Mosyle, Pi-hole, AdGuard, BrowserOS, Browser Use, goose, LangChain, CrewAI, Glama, Devin, Elicit, Otter.ai, AiSDR, Bardeen and Lindy are trademarks of their respective owners. Gartner is a registered trademark of Gartner, Inc.
Names are used only to identify the products discussed. AI Agent Blocklist is not affiliated with, sponsored or endorsed by these companies.