DNS block for AI agents

DNS Block AI Agents and AI Browsers on Your Network

Staff are installing AI browsers and agents that log in, click and send on their behalf. Your resolver already sees every name those agents look up, so it is the fastest place to say no.

This page shows what a DNS block stops, where it leaks, and how to load a daily AI agent DNS blocklist into the resolver you run today.

What your resolver answers
manus.imblocked
browseros.comblocked
smithery.aiblocked
office.comallowed

Blocked names get no usable address, so the agent never reaches its service.

Next step

AI Agent Blocklist: DNS-ready files

2,600+ AI agent and AI browser domains as hosts, domain and DNS filter import files, each with a risk grade, updated every day.

From $99 a month for one organisation

Why AI agents need a DNS block now

A chatbot answers questions. An agent acts: it opens pages, fills forms, reads your inbox and sends things. Every one of those agents depends on its maker's servers, and every connection starts with a DNS lookup.

1

User installs

An AI browser, an agent extension or a desktop agent goes on a work laptop. No ticket, no review.

2

Agent signs in

It uses the browser's live sessions: email, CRM, cloud storage, admin panels.

3

Agent calls home

Page content and instructions travel to the agent's own service to decide the next click.

4

DNS lookup

Before any of that, the device asks your resolver for the service's address. This is your control point.

What a DNS block stops

  • AI browsers reaching the agent backend that does the planning and clicking
  • Agent extensions sending page content to their service
  • Downloads of agent apps from their own sites
  • Hosted MCP servers and agent platforms reached by name
  • Every device on the network at once, including ones you do not manage

What it cannot stop alone

  • AI features served from domains you must allow, such as search or office suites
  • Devices that send DNS somewhere else (encrypted DNS, VPNs, relays)
  • Laptops off your network, unless the DNS filter follows them
  • Apps already installed: they stay on disk and retry
  • A specific path on an allowed domain, because DNS only sees names

What goes wrong when agents go unblocked

The risk is not that someone asks a question. It is that software with the user's logins acts on its own, on pages and instructions nobody checked.

Hidden instructions

Researchers showed in 2025 that hidden text in a web page comment could make an AI browser read a one-time code from the user's open email and send it out.

Quiet data copies

In 2025 a fake MCP server package copied every email sent through it to an outside address. Agents trust their tools.

Policy and audit gaps

Gartner's December 2025 advice was titled "Cybersecurity Must Block AI Browsers for Now". Auditors now ask what you did about it.

Who is most exposed

Finance and HR staff whose browsers are signed in to payment, payroll and personal data systems.

Developers who add coding agents and MCP servers that can run commands and read files.

Support and sales teams who try agents that read CRM records and send emails for them.

Admins themselves, whose sessions reach consoles that control the whole company.

Signs AI agents are already on your network

Your DNS logs are an inventory you already own. Before you block, look at what is being resolved. These are the patterns we see admins find first.

Lookups for AI browser update and sync services from a handful of laptops, every few minutes, all day.

Agent platform domains resolving at night or over weekends, when the user is not at the keyboard.

Names of MCP hubs and registries from developer machines that never appeared in your approved tools list.

Lookups to the services behind AI sidebar and agent extensions, from browsers you thought were locked down.

Requests for installer and download hosts of AI browsers shortly after a launch or price change in the news.

A drop in queries from some devices to your resolver at all: a hint that they moved to encrypted DNS.

Run the list in log-only mode for a week if your resolver supports it. You get a clear picture of who uses which agent before anyone sees a block. See also our guide to shadow AI agents.

Where DNS blocking of AI leaks, and how to close each gap

A DNS block is only as strong as the path that forces devices to use your resolver. Close these gaps first or the list will look like it does nothing.

Encrypted DNS in the browser

Chrome and Edge can send DNS over HTTPS. Set the DnsOverHttpsMode policy to off, or point it to your own resolver. In the Google Admin console the setting is called DNS-over-HTTPS.

Firefox default DNS over HTTPS

Answer the canary domain use-application-dns.net with NXDOMAIN. Firefox then keeps using the network resolver unless the user chose otherwise. Enforce it with Firefox enterprise policies.

Apple iCloud Private Relay

Return NXDOMAIN for mask.icloud.com and mask-h2.icloud.com to tell Apple devices the network does not allow the relay.

Outbound DNS to anywhere

On the firewall, allow port 53 and 853 only to your resolvers. Everything else is dropped, so a hard-coded public resolver fails.

Laptops off the network

A home or hotel network uses its own DNS. Use a roaming DNS client, an always-on VPN or device controls through Intune and Microsoft Defender.

AI inside allowed domains

AI Mode in Google Search or AI inside messaging apps shares domains with things you need. Switch those off with policy, for example the Chrome AI policies.

Block Meta AI, Google AI and other big-platform AI with DNS

Searches for "block meta ai dns" and "block google ai dns" usually hit the same wall: the AI feature lives on the same domains as the product you need. Here is what DNS can and cannot do for each.

ServiceOwn host name you can blockDNS alone?What to add
Meta AI webmeta.aiYesNothing for the standalone site.
Meta AI in WhatsApp, Instagram, FacebookNone: same domains as the appsNoBlock or allow the apps as a whole on managed devices.
Gemini app on the webgemini.google.com, gemini.googleYesTurn the Gemini app off in the Google Admin console for managed accounts.
Other Google AI toolsnotebooklm.google, opal.google, labs.googleYesCheck the service settings in the Admin console as well.
AI Mode and AI answers in Google SearchNone: same domain as searchNoChrome policy AIModeSettings for the address bar, plus browser controls.
Gemini in Chrome and its agent featuresNone you should rely onPartlyChrome policies GeminiSettings and GeminiActOnWebSettings. See disable Gemini in Chrome.
Standalone AI browsers and agentsTheir own service domainsYesRemove the apps from managed devices as well.

Host names change. Our files track the names each AI service uses and update every day, so a new host for an existing agent reaches your resolver without you hunting for it.

How we solve it: an AI agent DNS blocklist that stays current

Writing your own list means chasing hundreds of new agents, extensions and MCP hubs a month. The AI Agent Blocklist does that for you and ships in the formats DNS servers read.

2,600+

AI agent domains: AI browsers, browser agents, autonomous agents, agent platforms, MCP servers, coding agents, deep-research, meeting, voice and sales agents.

5,300+

Agent identifiers on devices: extension IDs, bundle IDs, Windows package IDs and process names, for the controls DNS cannot do.

Daily

New AI browsers, agents, extensions and MCP servers are added as they appear. Every entry is checked before it enters the list.

Plain domain listmanus.im
browseros.com
smithery.ai
Hosts file0.0.0.0 manus.im
0.0.0.0 browseros.com
0.0.0.0 smithery.ai
Adblock-style DNS rules||manus.im^
||browseros.com^
||smithery.ai^

Each entry also comes in CSV and JSON with the tool name, category, identifier type and risk grade, so you can build your own subsets.

How to block DNS lookups for AI agents, step by step

The steps are the same for most resolvers. The menu names differ; check the current admin console of yours.

  1. Decide the scope

    All agents, or only the higher-risk grades. Start with everything for guest and student networks, and the higher grades for staff if you have approved tools.

  2. Pick the file for your resolver

    Hosts format for resolvers that read hosts files, plain domains for list subscriptions, adblock-style rules for filters that support them.

  3. Load it as its own list

    Keep the AI agent list separate from malware and ad lists. You can then see its hits, switch it per group and add exceptions cleanly.

  4. Add your exceptions

    Approved AI tools go on an allow rule that sits above the blocklist. Write down who approved each one and why.

  5. Schedule the daily refresh

    Set the list to update once a day. Many resolvers default to weekly, which misses new agents for days.

  6. Close the leaks

    Turn off browser DNS over HTTPS, answer the canary domains, and block outbound DNS except to your resolvers.

  7. Test from a client

    Look up a listed name from a managed laptop and from a guest device. Both should get the block answer your resolver uses.

  8. Watch the logs

    Hits per device tell you who to talk to and which agents need a device-level block as well.

Load the list into common DNS servers

Use the format your server reads best. These are the usual routes; exact menus change between versions, so check the current documentation of your server.

Pi-hole and AdGuard Home

Add the list address as a subscribed blocklist and refresh daily. Step-by-step menus are on our Pi-hole and AdGuard Home page.

Cloud DNS filters

Import the domain list as a custom destination or block list. For one common service see Cisco Umbrella.

BIND with a response policy zone

Turn the domain list into an RPZ zone file where each name and its wildcard point to NXDOMAIN. Reload the zone after each daily update.

Unbound

Generate one local-zone line per domain with the always_nxdomain type in an included file, then reload. A local zone covers its subdomains too.

Windows Server DNS

Use DNS query resolution policies (Windows Server 2016 and later) that deny the listed names, created from the domain file with PowerShell. Check the current Microsoft documentation for syntax.

Firewalls with DNS filtering

Many firewalls read an external domain list by URL. See blocking AI agents at the firewall.

Which block answer to return

Your resolver can answer a blocked name in several ways. The choice changes what the user sees and how fast the agent gives up.

AnswerWhat the device seesGood forWatch out for
NXDOMAINThe name does not existAgents and apps: they fail fast and stop retrying soonerUsers get a plain browser error with no explanation
0.0.0.0 or null addressA name that points nowhereHosts files and most list-based filtersSome apps retry often, which fills your logs
Block page addressYour server answers insteadTelling people why and where to ask for an exceptionHTTPS sites show a certificate warning, not your page
REFUSEDThe resolver declines to answerRare cases where you want a hard stopSome clients then try another resolver

Whatever answer you choose, tell staff where AI requests go. A short intranet note with the approved tools and an exception form cuts tickets and stops people hunting for a way around the block.

What to block, what to watch, what to allow

A blunt block of every AI service breaks work and creates pressure to get around it. Use the category and risk grade in the list to set three tiers.

Block

Agents that act with the user's sessions and send data out.

  • AI browsers and browser agents
  • General autonomous agents
  • Hosted MCP hubs nobody approved
  • Higher risk grades in any category

Watch

Log, review weekly, decide per team.

  • Coding agents on developer subnets
  • Meeting and voice agents
  • Deep-research agents
  • Agent-building platforms

Allow

Tools you approved, with a named owner.

  • Your licensed AI assistant
  • Approved coding tools
  • Internal agents on your own domains
  • Exceptions with an end date

DNS is the network half. For the device half, block the browsers and extensions themselves: see how to block AI browsers and AI browser extensions.

DNS block lists for AI and for malware: run both

People searching "dns to block malware" and "dns block list" often already run a threat list. The AI agent list sits next to it and does a different job.

Malware and phishing listAI Agent Blocklist
What it targetsDomains used to attack youLegitimate AI agent services you have not approved
Why blockThe domain is harmfulThe agent acts with user logins and sends data out
ExceptionsRareCommon: approved tools per team
GranularityBlock allBy category and risk grade
Who decidesSecurity teamSecurity, IT and the business owner of each tool

Put the AI agent list on your resolver this week

One subscription covers every DNS server you run, plus the device files for the gaps DNS leaves.

  • Plain domain list, hosts file and adblock-style DNS rules
  • DNS filter import files and firewall external list format
  • CSV and JSON with tool name, category and risk grade
  • Extension, app and process identifiers for device controls
  • Updated every day; download access set up within one business day of an order

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.

DNS and AI blocking: questions admins ask

Can DNS block AI agents and AI browsers?

Yes, for the services they depend on. An agent must resolve its service domains before it can work, and a block answer stops it there. DNS does not remove installed apps, so pair it with device controls.

Is there an AI DNS blocklist I can import?

The AI Agent Blocklist ships 2,600+ AI agent domains as a plain domain list, hosts file, adblock-style list and DNS filter import files. Each entry has a category and a risk grade. The files update every day.

Can I block Meta AI with DNS?

You can block the standalone meta.ai site. Meta AI inside WhatsApp, Instagram and Facebook uses the same domains as those apps, so DNS cannot remove it without blocking the apps.

Can I block Google AI Mode or Gemini with DNS?

Standalone Google AI services with their own names, such as gemini.google.com, can be blocked. AI Mode in Search shares the search domain, so use the Chrome AIModeSettings and GeminiSettings policies for that.

Why does my DNS block not work in some browsers?

The browser is probably using DNS over HTTPS to a public resolver. Turn it off with the DnsOverHttpsMode policy, answer the Firefox canary domain with NXDOMAIN, and block outbound DNS except to your resolvers.

Can the same resolver block malware and AI agents?

Yes. Keep the AI agent list as its own list or group so you can see its hits, change it without touching malware blocking, and add exceptions for approved tools.

Meta and Meta AI are trademarks of Meta Platforms, Inc. Google, Gemini, Chrome and NotebookLM are trademarks of Google LLC. Microsoft, Windows, Windows Server, Edge, Intune and Microsoft Defender are trademarks of Microsoft Corporation.

Apple and iCloud Private Relay are trademarks of Apple Inc. Firefox is a trademark of the Mozilla Foundation. Manus, BrowserOS, Smithery, Pi-hole, AdGuard Home, BIND, Unbound and Cisco Umbrella 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.