Why banks, retailers and enterprises are blocking AI agents
Several large banks, online marketplaces and enterprise software makers now set limits on AI agents that log in or act for a user. Their reasons apply just as much to the agents your own staff run.
Where the rules are tightening
Next step
AI Agent Blocklist
Apply the same rule inside your company: decide which agents may act for your staff, then block AI browsers, agent extensions and agent platforms on your network and devices.
From $99 a month for one organisation
The trend: where AI agents now meet rules
Organisations that run accounts, payments and business data have started to write rules for AI agents. The details differ by sector, but the direction is the same: reading is often fine, acting needs permission.
-
Banking
Several large banks restrict AI agents on their sites
Several large banks now say third-party AI agents and assistants may not log in to customer accounts or make payments on their digital banking platforms. Limited read-only data sharing through approved open-banking providers continues.
For you: finance and payables are the first teams to check, because they work in bank portals every day.
-
Online marketplaces
Large marketplaces set rules for shopping agents in their user terms
User terms now spell out when buy-for-me agents, bots driven by language models and fully automated checkout are allowed, usually only with the marketplace's prior permission. Robots rules list the AI bots a site does not want.
For you: a purchasing team that lets an agent shop on a supplier marketplace may break the terms of the company account.
-
Enterprise software
Software makers route agents through approved interfaces
Makers of ERP, CRM and collaboration software publish API terms that say how AI agents may connect. Agent traffic is expected to use approved interfaces and agent gateways, not screen reading or unendorsed integrations.
For you: check the current API terms and your contract before your teams connect an agent to ERP or CRM data.
-
Websites and hosting
AI bots blocked by default on many websites
Large network and hosting providers now give site owners a one-click setting to block AI bots, and at least one made blocking the default for new sites in 2025.
For you: this protects websites. It does nothing about the agents your own staff run.
-
Online services in general
Terms against shared passwords and automated access
Most online banking, payroll and SaaS terms already forbid sharing passwords and automated access without permission. An agent that signs in with a person's session runs straight into those clauses.
For you: read the access clauses of the services your company relies on most.
-
Security analysts
Gartner: block AI browsers for now
Gartner's advisory "Cybersecurity Must Block AI Browsers for Now" (December 2025) told security leaders to block AI browsers in their organisations for the time being. We summarise it on our page about Gartner's advice to block AI browsers.
The six reasons behind every block
Banks, retailers and software vendors use different words, but the reasons repeat. Read them as a checklist for your own company, because each one also applies to an agent that acts inside your accounts.
1. Nobody knows who acted
The agent signs in with a person's password or session. Logs show the person, so the bank or the vendor cannot tell a human click from an agent's click.
2. Liability is unclear
When an agent buys the wrong thing, pays the wrong invoice or changes a setting, someone has to carry the loss. Contracts were written for people, not for software acting on their behalf.
3. Credentials leave the owner
Agents need logins, tokens or saved sessions to act. Handing them over breaks the basic rule of most online banking and SaaS terms: do not share your password.
4. Fraud controls are built for people
Fraud checks look at devices, speed and behaviour. An agent inside a normal browser session looks like the customer, so controls that would stop a bot never fire.
5. Data leaves controlled paths
An agent that reads screens or calls APIs on its own can copy pricing, customer and business data in bulk. Vendors want that traffic on routes they can see, meter and secure.
6. Agents can be steered
Researchers showed in 2025 that text hidden in web pages and emails can redirect AI browser agents. A steered agent acts with the user's full rights. Read more about prompt injection in AI browsers.
The same reasons apply inside your company
A bank blocks agents that act inside a customer's account. Your company has the same exposure, because staff run agents inside company accounts every day. This table maps each reason to what it looks like on your network.
| What the bank or vendor worries about | The same risk in your company | What to do |
|---|---|---|
| Agents log in with customer credentials | Staff let an AI browser use their signed-in sessions in email, CRM, finance and admin consoles | Block AI browsers and agent extensions on managed devices |
| Nobody can tell person from agent | Your audit logs show the employee for every action the agent took | Allow only agents with their own identity, block the rest |
| Agents move money or buy things | Accounts payable or purchasing staff use agents on bank portals and supplier sites | Strictest rules for finance and purchasing groups |
| Terms forbid automated access | Your bank, marketplace or ERP terms may forbid the agent your staff use | Block agents on those services, document the rule |
| Bulk data leaves through unapproved routes | Agent platforms connected to drives, mailboxes and ERP data copy it to outside services | Block unapproved agent platforms and MCP servers at the network |
| Agents can be steered by page content | One web page or email with hidden instructions can steer a staff member's agent inside your systems | Block acting agents, keep read-only assistants |
Who in your company is exposed
Exposure follows the accounts an agent can reach. These six groups touch the services where banks, marketplaces and software makers now set rules for agents.
Bank portals, payment runs and card programmes. An agent here can move money with a person's rights.
Marketplaces and supplier portals, where user terms now set rules for buy-for-me agents.
Agents that call ERP or data platform APIs outside the routes the vendor endorses.
Agents with access to mailboxes, calendars and approvals act with senior rights.
Coding agents run in a shell with keys, tokens and production access.
Browser agents in CRM and help desk tools reach customer data at scale.
Signs that agents already act for your staff
Most companies find agents before they write a rule about them. Look for these signs in your device inventory, browser reports, identity logs and DNS logs. Our guide to shadow AI agents covers each source in more detail.
- AI browser apps such as Comet or ChatGPT Atlas in the installed software report
- Agent extensions in Chrome or Edge that can read and change data on all websites
- Consent grants in your identity provider to agent platforms that read mail or files
- DNS lookups to agent platforms, agent browsers and MCP services
- Sign-ins from one user that click through forms faster than a person can
- Warnings from your bank or a supplier about automated access to your account
- Expense lines for agent subscriptions paid on personal or team cards
- Coding agent tools installed by developers without a request
The lesson for IT: decide which agents may act for your staff
Banks, marketplaces and software makers all answered one question first: which agents may act, for whom and where. Answer the same question for your company, write it down, then enforce it.
Which agents may act?
Name the agents you approve, each with an owner and a purpose. Every other acting agent is blocked by default, so a new tool starts with a request, not a cleanup.
For whom?
Finance, purchasing, executives and developers carry different risks. Give each group its own list instead of one rule for everyone.
On which services?
Bank portals, supplier marketplaces and ERP systems have their own terms for agents. Where a service says no, your rule says no too.
With whose identity?
An agent that borrows a person's session hides in your logs. Prefer agents with their own identity, scoped rights and a clear audit trail.
Who is accountable?
Decide in advance who answers for a payment, order or message an agent made. In most companies it is the person who started the task.
How is it enforced?
A written rule alone does not stop an install. Back it with block lists at the network, in browsers and on devices, and review the hits every week.
Check your own service terms first
Before you approve any acting agent, read what your key providers say about automated access. These are the clauses to look for.
- Online banking: sharing of credentials and third-party access to payments
- Payroll and HR platforms: automated sign-ins and bulk exports
- Supplier marketplaces: buy-for-me agents and automated checkout
- ERP and CRM: API terms for AI agents and approved integration routes
- Cloud storage and mail: which outside apps may hold access tokens
- Company cards: who may use them and through which channels
- Audit requirements: what your auditors expect to see for automated actions
- Your own policies: whether your acceptable use policy already covers agents
How we solve it: the same rule, applied with the AI Agent Blocklist
Banks block agents that log in and act, and keep read-only access open. The AI Agent Blocklist gives you the files to do the same on your network, browsers and devices, updated every day.
2,600+ AI agent domains
AI browsers and browser agents, general autonomous agents, personal AI assistants, agent builders, multi-agent orchestration, MCP servers, autonomous coding agents, deep research agents, meeting and voice agents, AI sales agents and AI RPA.
5,300+ agent identifiers on devices
Chrome, Edge and Firefox extension IDs of agent extensions, macOS bundle IDs, Windows package IDs and process names of AI browsers and agent apps, and the command-line coding agents with their package names.
Six steps to put the rule in place
Write the rule down
State which agents may act, for whom, and where. Our AI agent acceptable use policy gives you a starting text.
Choose block all or block by risk
Every entry carries a category and a risk grade from Low to Severe. Block every acting agent, or start with the higher grades and widen later. Our guide on allowlists and blocklists for AI agents helps you decide.
Load the domains at the network
Add the domain list to your firewall as an external list or to your DNS filter as a block list. See blocking AI agents at the firewall for each platform.
Block agent extensions in browsers
Put the extension IDs into the Chrome and Edge extension block list policy. An example entry is Manus AI Browser Operator,
cecngibhkljoiafhjfmcgbmikfogdiko. Our guide to blocking AI browser extensions shows the policies.Stop AI browsers on devices
Use the macOS bundle IDs and Windows package IDs in Jamf or Intune. Comet, for example, is
ai.perplexity.cometon macOS andPerplexity.Cometin winget. Our page on blocking AI browsers covers the rest.Review and widen
Check your DNS and firewall logs for blocked agent lookups each week. Each hit is a person who needs a rule explained or an approved tool offered.
What to block, what to watch, what to allow
Banks drew the line between reading and acting. That line works for most companies too, and it keeps useful AI in place while acting agents stay out.
Block
- AI browsers on managed devices
- Agent extensions that act on pages
- General autonomous agents and agent platforms
- Unapproved MCP servers and agent builders
- Buy-for-me and AI RPA services
Watch
- DNS hits on agent domains from finance and purchasing
- New agent extensions in browser reports
- Consent grants to agent apps in your identity provider
- Coding agents on developer machines
Allow
- Approved chat assistants that only answer questions
- Agents registered with their own identity and an owner
- Read-only data sharing through approved providers
- Coding agents with locked permission settings
Stricter rules for the teams that need them most
You do not need one rule for everyone. Group your users and pick the risk grades each group may use. Separate files per group are easy to build from the CSV.
Block AI agents the way banks do
Pick the categories and risk grades you block, load one file per tool, and let the daily updates add new agents as they appear. After an order we set up your download access within one business day.
- Plain domain list, CSV and JSON with every column
- Hosts file, PAC file and firewall external list format
- DNS filter import files
- Chrome and Edge extension block lists for Group Policy, Intune and the Google Admin console
- macOS and Windows identifiers for Jamf and Intune
- Tool name, category, identifier type and risk grade on every entry
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 companies blocking AI agents
Why are banks blocking AI agents?
Several large banks now restrict third-party AI agents and assistants that log in to customer accounts or make payments on their digital banking platforms.
The core problem is identity: a bank sees the customer's credentials but cannot tell whether a person or an agent is acting. Limited read-only data sharing through approved open-banking providers continues.
Why do online marketplaces set rules for shopping agents?
Marketplaces want to keep control of accounts, checkout, pricing data and fraud checks. Their user terms now often say that buy-for-me agents and fully automated checkout need the marketplace's prior permission.
For a business buyer, this means an agent that shops on a supplier marketplace may break the terms of the company account.
Are enterprise software makers blocking third-party AI agents?
Mostly they set routes rather than block outright. Enterprise software makers publish API terms that say how agents may connect, usually through approved interfaces and agent gateways.
Check the current terms and your contract before your teams connect an agent to ERP or CRM data.
Why are websites blocking AI agents?
Sites block agents to keep control of accounts, checkout, pricing data and terms of use. Robots rules, terms of service and bot settings at the network edge are the tools they use.
Each of those protects the website. None of them decides which agents your own staff may run.
Why do so many websites block AI bots by default?
Large network and hosting providers now give site owners a one-click setting to block AI bots, and at least one made blocking the default for new sites in 2025.
That change protects websites from bots. It does not stop your own staff from running AI agents.
Should my company block AI agents too?
Most companies should block unapproved acting agents, such as AI browsers and agent extensions, and allow a short list of approved assistants.
The same reasons banks give apply to you: agents act with a person's identity, service terms may forbid them, and your logs cannot tell the person from the agent. Start with our guide to blocking AI agents.
Can we still allow read-only AI assistants?
Yes. Banks allow read-only access through approved providers while they restrict agents that log in and transact, and you can do the same. Block the agent browsers, agent extensions and agent platforms that act, and keep approved chat assistants that only answer questions.
Gartner is a trademark of Gartner, Inc. Perplexity and Comet are trademarks of Perplexity AI, Inc. ChatGPT Atlas is a trademark of OpenAI. Manus is a trademark of its owner.
Chrome and Google Admin are trademarks of Google LLC. Microsoft, Edge and Intune are trademarks of Microsoft Corporation. Jamf is a trademark of Jamf Software, LLC. FortiGate is a trademark of Fortinet, Inc.
Names are used only to identify the products discussed. AI Agent Blocklist is not affiliated with, sponsored or endorsed by these companies.