Cisco Umbrella guide

Cisco Umbrella: block generative AI agents, AI browsers and AI sites

The AI tools that worry security teams now act on their own: they browse, send and buy with your users' logins. Umbrella sees every lookup they make. A block destination list with the right domains stops them at DNS.

DNS policyBlock destination list of AI agent domains. Subdomains included automatically.
Web policySame list in a block rule, for users on the secure web gateway.
APIWeekly job adds new agents and removes retired ones.

Next step

AI Agent Blocklist for Cisco Umbrella

2,600+ AI agent domains as a plain list for destination list upload, plus CSV for the API, updated every day.

From $99 a month for one organisation

Block generative AI, or block the AI that acts?

"Cisco Umbrella block generative AI" can mean blocking every AI chat site. Most teams find that too blunt, because they pay for one assistant. The bigger risk is agents, which do things rather than answer questions.

AI that answers

  • Chat assistants and writing tools
  • Risk: users paste data into a prompt
  • Usually handled by an approved tool and a policy
  • Your Umbrella content categories stay in place; we do not describe them here

AI that acts

  • AI browsers such as Comet and ChatGPT Atlas
  • Cloud agents, agent extensions, MCP servers, coding agents
  • Risk: actions taken with the user's sessions and tokens
  • Gartner advised in December 2025 to block AI browsers for now
Our list covers the second column and adds it as a destination list next to everything you already run. For the reasoning behind blocking agents first, see AI agent security risks.

Signs AI agents already use your Umbrella identities

Umbrella logs every lookup per identity, so you can see agent use before you write a single rule. These are the patterns that usually show up in the first week of looking.

AI browser updates

Steady lookups for an AI browser's update and sync hosts from machines that should run only your standard browser.

Working-hours agents

Repeated lookups for agent platforms from finance, sales and support during the day, often from the same few users.

MCP traffic

Lookups for MCP directories and hosted MCP servers from developer subnets and build servers.

Meeting bots

Lookups for note-taker services around the times of customer calls.

Roaming laptops

The same patterns from roaming clients at home, where no office firewall would see them.

Student devices

AI browser and homework agent lookups from student networks and Chromebooks.

To find these, run an Activity Search in Umbrella reporting filtered on a handful of agent domains from our list. It shows which identities ask for them and how often.

DNS policy or web policy: where the list works

Umbrella destination lists come in two kinds, one for DNS policies and one for web policies. You choose the kind when you create the list, and it cannot be used in the other kind of policy.

DNS destination listWeb destination list
Entry typesDomains for block and allow; URLs for block lists only; IP addresses for allow lists onlyDomains, URLs, IPv4 addresses and CIDR ranges
SubdomainsA domain entry covers all its subdomains; wildcards are not usedDomain entries cover the domain's traffic through the gateway
SizeNo hard maximum; 5,000 entries per list recommendedMaximum 5,000 destinations per list
Takes effectWhen added to a DNS policyWhen added to a rule with action Block or Allow
Who is coveredEvery identity on the policy: networks, roaming clients, sitesUsers whose web traffic goes through the secure web gateway
Our AI agent domain list fits in a single list of either kind. If you also block other large lists, split by category into several destination lists to stay under the recommended size.

Step 1: create the block destination list

These steps follow the Umbrella documentation. The dashboard is updated often, so check the current admin console if a label moved.

  1. Go to Policies > Policy Components > Destination Lists and click Add.
  2. Name the list clearly, for example "AI Agent Blocklist".
  3. In This Destination List Type, choose DNS Policies and select Blocked. Create a second list for web policies later if you use the gateway.
  4. Add one domain by hand so the list can be saved, then save it.
  5. Bulk upload is not available while the list is being created. Open the saved list and use Upload for the rest.

Step 2: upload the AI agent domains

Upload takes a text file with one destination per line. Use the plain domain list from our download, which has exactly that shape.

Upload

  1. Expand the list in Destination Lists.
  2. Click Upload and choose the domain file.
  3. Umbrella checks each line and adds the valid ones.
  4. Lines it cannot use come back in a spreadsheet you can download, with the reason for each.

Format rules

  • No protocol in front of a domain
  • No wildcards: the domain covers its subdomains
  • No duplicates within one list
  • For allow lists, URLs are not supported; use the domain

Step 3: apply the list in DNS and web policies

A destination list does nothing until a policy uses it. Do the DNS policy first; it covers every identity, including roaming laptops with the Umbrella client.

DNS policy

  1. Open Policies > Management > DNS Policies and edit the policy for your users.
  2. In the destination lists step, select the AI Agent Blocklist.
  3. Save, then test from a device that uses this policy.
  4. For a company-wide block, the Global List applies to all DNS policies and identities; use your own list when only some identities should be covered.

Web policy

  1. Create a destination list of type Web Policies with the same domains.
  2. Open the web policy ruleset and click Add Rule.
  3. Give it a Rule Name and set the Rule Action to Block.
  4. Click Add Destination, choose the list and click Apply. Save.
Web destination lists can only be used in rules whose action is Block or Allow, not Warn. If you want a soft start, apply the DNS policy to a pilot identity group first, then widen it and switch on the web rule.

Allow lists win: how to keep approved AI working

Umbrella checks a policy's allow destination lists first. If a domain is on both an allow and a block list on the same policy, it is allowed. Use that to protect your approved tools.

Checked first

Allow lists

Your approved AI tools, such as the assistant you licensed.

Then

Security settings

Your existing security categories, unchanged.

Then

Block lists

The AI Agent Blocklist and your other block lists.

Then

Other settings

The rest of your policy, unchanged.

Keep the allow list short, with an owner and a review date for each entry. Allowing a whole parent domain also allows every subdomain under it, including agent services you meant to block. Our guide on allowlists and blocklists for AI agents helps you decide what goes where.

Keep the destination list current with the API

Umbrella does not pull destination lists from a URL. Our list changes every day, so a short weekly job keeps Umbrella in step without manual uploads.

TaskCall in the Policies API v2Note
Find the listGET /destinationlistsNote the list's ID
Add new domainsPOST /destinationlists/{id}/destinationsUp to 500 destinations per request
Remove retired domainsDELETE /destinationlists/{id}/destinations/removeUp to 500 IDs per request
Check the resultGET /destinationlists/{id}/destinationsCompare with our file
# weekly job, in plain steps 1. Download the new domain list 2. Compare it with last week's file 3. POST new domains in batches of 500 4. DELETE removed domains in batches of 500 5. Log the counts for your change record
The CSV in our download has the tool name, category and risk grade for every domain. Use it to build separate destination lists, for example one for AI browsers and one for MCP servers, so you can switch groups on and off.

What to put in which destination list

Splitting the list by category lets you block the riskiest groups at once and leave others in a monitored state while you talk to the teams who use them.

AI browsers and browser agents

Block now. Remove the apps with MDM as well; see blocking AI browsers.

Autonomous agents

Block now. Examples are manus.im and genspark.ai.

MCP servers and directories

Block, with approved servers on the allow list. Example: smithery.ai.

Coding agents

Block unapproved ones; allow the one engineering signed off.

Meeting and voice agents

Monitor first, then decide per team. Example: read.ai.

Deep-research agents

Monitor first; analysts may need an approved one.

Google AI Mode, Gemini and Umbrella

DNS sees host names, not pages. That matters for Google, which runs several AI features on the same hosts as Search.

FeatureBlockable at DNS?What to do
Gemini appYes, it has its own host, gemini.google.comAdd it to the block list if Gemini is not approved
Google AI ModeNo, it runs on google.comSet the Chrome policy AIModeSettings to disabled on managed browsers
Gemini in ChromeNot by DNS aloneChrome policies; see disabling Gemini in Chrome

What Umbrella cannot see, and what covers it

DNS blocking is broad and cheap, but some agent activity never makes a DNS query you control. Plan for these gaps from the start.

Gaps

  • Local MCP servers that run on the laptop
  • Extensions already installed, until they call home
  • Browsers using their own encrypted DNS
  • Devices without the Umbrella client off the network

Cover them with

Test it and read the results

Run a short test on one identity before you widen the policy. It takes ten minutes and confirms the list, the policy and the client.

  1. From a test device, look up an AI agent domain from the list. It should resolve to the Umbrella block page.
  2. Open the same site in a browser and confirm the block page appears.
  3. Open your approved AI assistant. It should load normally.
  4. In Umbrella reporting, run an Activity Search for the test identity and look for the blocked requests with your list name.
  5. Repeat from a roaming laptop off the office network.
  6. Record the result, the date and the list version in your change ticket, so the next review starts from a known state.
If a blocked domain still resolves normally, check three things in order: that the device really uses the policy you edited, that no allow list on that policy contains a parent domain, and that the browser is not using its own encrypted DNS.

A rollout plan for Umbrella

Destination lists block or allow; there is no watch-only mode for them. So stage the rollout by identity group and by category instead.

  1. Look first. Use Activity Search for a week to see which agent domains your identities use and who uses them.
  2. Pilot group. Apply the block list to a DNS policy for IT and one friendly team. Fix any approved tool that breaks with an allow list.
  3. High-risk categories for everyone. Block AI browsers, autonomous agents and MCP servers for all identities. Tell users what changed and where to ask.
  4. The full list. Add the remaining categories once teams with a real need have an approved tool on the allow list.
  5. Weekly upkeep. Run the API job, review new exceptions and remove expired ones.

Umbrella in schools and districts

Many districts run Umbrella on student networks and Chromebooks. AI browsers and homework agents raise the same issues there, plus a few of their own.

What districts see

  • Students install AI browsers on take-home devices
  • Agents complete assignments inside learning platforms with the student's login
  • Extensions that read every page, including grades and messages

What helps

  • The AI agent list on the student DNS policy, separate from staff
  • Chrome policies for extensions and AI features on managed Chromebooks
  • See AI browsers on Chromebooks

One list for DNS and web policies

Upload it once, keep it current with a weekly job, and let allow lists protect the AI tools you approved. We set up your download access within one business day of your order.

  • Plain domain list for upload
  • CSV and JSON with category and risk grade
  • DNS filter import files and hosts file
  • Extension and app identifiers for devices
  • Updated every day

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.

Cisco Umbrella and AI blocking: questions

How do I block AI sites in Cisco Umbrella?

Create a destination list under Policies, Policy Components, Destination Lists, set it to block and fill it with the AI agent domains from our list. Then add it to your DNS policy, and to a block rule in your web policy if you use the secure web gateway.

What is a Cisco Umbrella block list?

In Umbrella a block list is a destination list set to block. It holds domains, and for web policies also URLs and IP addresses. A list only takes effect once a policy uses it.

How do I block a website with Cisco Umbrella?

Add the domain to a block destination list that is used in the DNS policy for the identities you want to cover. A domain entry also blocks its subdomains, so you do not need wildcards.

Can Cisco Umbrella block generative AI without blocking approved tools?

Yes. Put the AI agent list in a block destination list and your approved tools in an allow destination list on the same policy. Allow destination lists take precedence over block destination lists.

How many domains fit in one Umbrella destination list?

Umbrella documents a maximum of 5,000 destinations for a web destination list. DNS lists have no hard maximum, with 5,000 per list recommended. Our AI agent domain list fits in one list.

Can Umbrella block Google AI Mode?

Not at DNS level, because AI Mode runs on google.com and DNS sees only the name. The Gemini app has its own host and can be blocked. For AI Mode in Chrome, use the AIModeSettings policy; see Chrome AI policies.

Do roaming laptops get the AI block outside the office?

Yes, if the Umbrella roaming client or another Umbrella connector is running on them. The DNS policy applies to the identity wherever it is, so the AI agent list follows the laptop home. Devices without the client are only covered while they use a protected network.

How do I keep the Umbrella destination list up to date?

Use the destination lists API to add new entries and remove retired ones, in batches of up to 500 per request. Most teams run it weekly. A manual upload of a text file also works.

Cisco and Cisco Umbrella are trademarks of Cisco Systems, Inc. Google, Chrome and Gemini are trademarks of Google LLC. Perplexity and Comet are trademarks of Perplexity AI, Inc. ChatGPT and ChatGPT Atlas are trademarks of OpenAI.

Manus, Genspark, Smithery and Read AI are trademarks of their respective owners. Gartner is a 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.