Claude in Chrome

Claude in Chrome at work: risks and how to block it

Claude in Chrome is an extension that reads, clicks and types in your staff's logged-in tabs. On some business plans it is switched on by default, and on personal accounts you have no say at all.

Next step

AI Agent Blocklist: block Claude in Chrome and every other agent extension

Extension IDs of agent extensions for Chrome, Edge and Firefox, ready for Group Policy, Intune, Jamf and the Google Admin console.

From $99 a month for one organisation

What Claude in Chrome is and what it can reach

Anthropic's extension opens a side panel in Chrome. You give it a task, and it works through the page: reading, scrolling, clicking, filling in forms and opening other tabs.

Extension ID

fcoeoabgfenejglbffodgkkbkcdhcgfn in the Chrome Web Store

Timeline

Research preview for some Max users in August 2025, then opened to all paid plans in December 2025

What it sees

Screenshots of the active tab, including anything confidential on it

What it uses

The sessions the user is already signed in to in that browser profile

It also connects to other Claude products

  • On several plans the side panel runs as a Claude Cowork session
  • Users with Claude Desktop can start desktop tasks that work in the browser
  • Side panel sessions are saved to history and can be reopened on other devices

Defaults that matter to admins

  • Team plans: enabled by default
  • Enterprise: on unless an admin already disabled it
  • Enterprise with HIPAA: off until an owner turns it on
  • Enterprise users are not told automatically when it is enabled

Extension agent, AI browser or desktop agent: why the form matters

The same kind of agent can arrive in three forms. Each one needs a different control, and an extension is the easiest of the three to miss.

FormHow it arrivesWhat stops it
Agent extension, such as Claude in ChromeOne click in the Chrome Web Store, no admin rights, inside the browser you already manageExtension block or allow lists in browser policy
AI browserA separate app installed next to your managed browserApp blocking in MDM, plus domain rules
Desktop or computer-use agentAn installed app that operates the whole screenApp blocking, OS privacy permissions and dedicated machines; see computer use agents

Is Claude in Chrome safe? The risks Anthropic itself lists

Anthropic is open about the limits. Its safety guidance names the risks below, and its own advice is to start with trusted sites and keep the extension away from sensitive accounts.

RiskWhat it means at workWho is exposed
Prompt injectionInstructions hidden in a site, email or document can lead the agent into actions the user did not ask for. Anthropic says this risk is reduced, not removed.Anyone who points it at mail, tickets or public pages
Sensitive data on screenEverything visible in the controlled tab is captured in screenshots and becomes part of the conversation.Finance, HR, support, anyone with customer records open
Regulated dataAnthropic recommends against using it on pages with regulated data. For HIPAA organisations it is off by default and not covered by the BAA.Healthcare, schools, public sector
No zero data retentionAnthropic's admin documentation states zero data retention is not supported for the extension.Teams with strict retention contracts
User responsibilityAnthropic says the user remains responsible for what the agent does, including messages sent and data changed.Every user, and the company behind them
Personal accountsA personal plan on a work laptop gets none of your organisation's site lists or switches.Every managed device without an extension policy

Hidden-instruction attacks on browser agents are explained on our page about prompt injection in AI browsers.

The safeguards built in, and where they stop

Permission modes users choose

  • Manually approve: asks before each action
  • Automatically approve: checks each action for safety and pauses when needed; the default in the Cowork side panel
  • Skip all approvals: no pauses and no automatic check, apart from connector tools set to always ask

Rules that apply in any mode

  • Always asks before changing permissions, granting authorizations, entering sensitive data or permanent deletions
  • Refuses purchases, account creation, handling card or ID data and investment trades
  • Blocks categories such as adult and known pirated content, and asks before financial sites

Why this is not enough on its own: the user picks the mode, and Skip all approvals removes the pauses. Built-in rules protect against obvious harm, not against a convincing instruction inside a page your user trusts. Company policy has to decide who may use it at all.

Control it from two sides: Claude's settings and the browser

Claude's organisation settings govern accounts in your Team or Enterprise organisation. Browser policy governs the extension on your managed devices, whatever account signs in. Use both.

In Claude (Team and Enterprise)

  1. Sign in as an Owner or Primary Owner and open Organization settings > Claude in Chrome.
  2. Switch off Enable for your team to turn it off for the organisation.
  3. Or keep it on and set a restrictive site allowlist plus a blocklist. The same list governs Cowork's built-in browser.
  4. On Enterprise, use the per-role capability in custom roles; it is separate from Cowork access.
  5. To stop Claude Desktop from driving the browser, switch the extension off or disable isLocalDevMcpEnabled in the Enterprise configuration.
+

In the browser (any account)

  1. Block the extension ID for everyone who is not approved.
  2. For approved users, deploy it through the Google Admin console or your MDM rather than self-install.
  3. Set the managed extension policy forceLoginOrgUUID so the extension works only with your Claude organisation.
  4. Keep personal profiles out with browser sign-in and profile policies.

What blocked sites look like: Anthropic documents that if a user types a blocked URL in Cowork's built-in browser, the page loads with a banner saying the site is blocked by the organisation's policy and Claude's tools are disabled on that page.

How to block the Claude Chrome extension with policy

The extension ID is the same on every device. Pick the route that matches your fleet. A blocked extension that is already installed is disabled and the user cannot switch it back on.

Google Admin console

  1. Go to Devices > Chrome > Apps & extensions > Users & browsers.
  2. Select the organisational unit or group.
  3. Find the extension and set Installation policy to Block, then Save.
  4. Or block it from Reports > Apps & extensions usage, which also removes it from browsers in that unit.

Full Google Admin guide

Group Policy or registry

  1. Load the Chrome administrative templates.
  2. Add the ID to the Chrome extension install blocklist policy.
  3. The registry location is SOFTWARE\Policies\Google\Chrome\ExtensionInstallBlocklist, one numbered string value per ID.
  4. Run gpupdate /force and check chrome://policy on a test PC.

Group Policy and registry guide

Microsoft Intune

  1. Create a Settings catalog profile for Windows.
  2. Add the Google Chrome extension blocklist setting and enter the ID.
  3. Assign to device or user groups and sync a test machine.

Intune extension guide

Jamf on Macs

  1. Create a configuration profile for the Chrome preference domain.
  2. Set ExtensionInstallBlocklist as an array containing the ID.
  3. Scope to the Macs and verify in chrome://policy.

Jamf extension guide

Remove it everywhere with ExtensionSettings

The ExtensionSettings policy accepts an installation mode per extension. blocked stops installs; removed also uninstalls existing copies. You can add a message users see when an install is blocked.

{
  "fcoeoabgfenejglbffodgkkbkcdhcgfn": {
    "installation_mode": "removed",
    "blocked_install_message": "Browser agents need IT approval. Ask the service desk."
  }
}

Test the block in five minutes

Policy arrived

chrome://policy lists the blocklist or ExtensionSettings value with status OK on the test device.

Install refused

The store page refuses the install, and shows your custom message if you set one.

Old copy gone

An existing copy is disabled with the blocklist, or removed when you use the removed mode.

Edge too: Edge can install extensions from the Chrome Web Store when allowed. If you permit that, add the same ID to Edge's extension blocklist as a precaution. See the Edge extension blocklist.

Allow it with limits instead of a full block

Some teams have a good reason to use it. If you approve it, narrow what it can touch so one bad page cannot reach your most sensitive systems.

Allowlist in Claude

List the sites the agent may act on. Anthropic recommends starting restrictive during rollout.

Runtime blocked hosts

In ExtensionSettings, runtime_blocked_hosts stops an extension interacting with listed sites, up to 100 entries. Use it for banking, HR and admin consoles.

Separate profile

Anthropic advises a separate browser profile without banking, health or government accounts. Make that profile the only one where the extension is installed.

Running an allowlist for all extensions, not just this one, is covered in building a Chrome extension allowlist.

Is it already installed? Where to look

Check before you block, so you know who will notice. These sources show installs by extension ID, which is the only reliable way to find it.

SourceWhereWhat you learn
Google Admin consoleDevices > Chrome > Reports > Apps & extensions usageInstall counts across managed browsers, profiles and ChromeOS devices, and the permissions requested
Microsoft DefenderVulnerability management > Inventories > Browser extensionsWindows devices and users with the extension, and whether it is turned on
Claude organisation settingsOrganization settings in your Team or Enterprise organisationWhether the feature is on for your organisation and which sites are allowed
A test devicechrome://extensions and chrome://policyWhether your block policy has arrived and applies

If you approve it: a careful rollout in four steps

1. Block for everyone first

  1. Put the ID on the browser blocklist for all users.
  2. Tell staff why, and how to request access.

2. Open a small pilot

  1. Allow it for one group through a separate organisational unit or policy.
  2. Set forceLoginOrgUUID and a short site allowlist in Claude.

3. Watch and adjust

  1. Review which sites the pilot needed and which it tried to reach.
  2. Add sensitive systems to runtime blocked hosts and Claude's blocklist.

4. Decide and record

  1. Expand, keep the pilot, or remove it.
  2. Record the owner, scope and next review date in your agent inventory.

Do not skip step one. Approving the extension for a pilot while it stays open for everyone else means the pilot rules never apply to most of your users.

Why blocking claude.ai is the wrong tool for this job

What a domain block does

  • Stops every use of Claude on your network, including approved chat and coding use
  • Breaks the extension only while the device is on your network
  • Leaves the extension installed and ready on home Wi-Fi

What an extension block does

  • Removes only the browser agent and leaves approved Claude use alone
  • Works on every network, because the browser enforces it
  • Shows up in policy reports as evidence of control

A domain block also tells you little. Your logs show traffic to Claude, but not whether it came from approved chat, a coding tool or the browser agent. An extension block answers that question by design.

Use domain blocks for organisations that do not allow Claude at all. Coding use has its own controls, covered in blocking Claude Code.

How we solve it: one block list for every agent extension

Claude in Chrome is one agent extension among many. Block it alone and the next one fills the gap. The AI Agent Blocklist keeps the whole category current: every entry is checked before it enters the list, and the lists update every day.

What you getFormatWhere it goes
Agent extension IDs for Chrome and EdgePolicy-ready block listsGroup Policy, Intune, Google Admin console, Jamf
Agent extension IDs for FirefoxID listFirefox enterprise policies
AI browser and agent app identifiersmacOS bundle IDs, Windows package IDs, process namesJamf, Kandji, Mosyle, Intune
Agent domains with risk gradesDomain list, CSV, JSON, firewall and DNS formatsDNS filter, firewall, secure web gateway

Start with the category approach on blocking AI browser extensions. If Chrome's own built-in AI is on your list too, see disabling Gemini in Chrome.

Block browser agents before they reach your tabs

  • 5,300+ agent identifiers on devices, including Chrome, Edge and Firefox extension IDs
  • 2,600+ AI agent domains, each with category and risk grade
  • Policy files for Group Policy, Intune and the Google Admin console
  • Updated every day; download access set up within one business day of your order

Everything in the download: the AI Agent Blocklist.

From $99 a month for one organisation See plans Ask a question

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.

Claude in Chrome: safety and blocking FAQ

Is Claude in Chrome safe for work?

It can be used with care, but it acts inside logged-in tabs and sees what is on screen. Anthropic says prompt injection risk is reduced, not removed, and advises against regulated data. Many organisations block it by default and allow reviewed roles only.

What are the main Claude in Chrome risks?

Hidden instructions in pages or mail that steer the agent, screenshots of sensitive pages entering the conversation, actions taken with the user's sessions, and personal accounts on work devices that admins cannot control.

How do I block the Claude Chrome extension?

Add fcoeoabgfenejglbffodgkkbkcdhcgfn to Chrome's extension blocklist through Group Policy, Intune, Jamf or the Google Admin console, or set it to removed in ExtensionSettings. Team and Enterprise owners can also switch it off in Claude's organisation settings.

Why does Claude in Chrome show as disabled?

Usually because an admin blocked it. Chrome disables a blocked extension that is already installed, and the user cannot turn it back on. On business plans an owner may also have switched it off for the organisation.

How do we uninstall Claude in Chrome from all devices?

Set installation_mode to removed for the ID in ExtensionSettings. Chrome removes it and stops reinstalls. In the Google Admin console you can also block it from the apps and extensions usage report.

Does Claude in Chrome have an allowlist?

Yes. Team and Enterprise owners can set a site allowlist and blocklist under Organization settings > Claude in Chrome. The same list governs Cowork's built-in browser. Anthropic recommends starting restrictive.

Why does Claude in Chrome refuse some sites?

Anthropic blocks some categories by default, such as adult and known pirated content, and asks before financial sites. Your organisation's blocklist adds more, and some sites refuse automated agents on their own.

Can we allow Claude in Chrome only with our company account?

Anthropic documents a managed extension policy, forceLoginOrgUUID, that limits which Claude organisation the extension works with. Combine it with your organisation settings and block the extension for everyone not approved.

Anthropic, Claude, Claude in Chrome, Claude Code, Claude Desktop and Cowork are trademarks of Anthropic, PBC. Google, Chrome, Chrome Web Store, Gemini and Google Admin are trademarks of Google LLC. Microsoft, Windows, Intune and Edge are trademarks of Microsoft Corporation.

Apple is a trademark of Apple Inc. Firefox is a trademark of the Mozilla Foundation. Jamf, Kandji and Mosyle 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.