topicsthe AI3 numbers2 guides

AI browser extensions

AI extensions read page content and typed text with no audit trail. What they access, and how an extension policy holds.

in one lineAI browser extensions can read open tabs, typed text, and clipboards, and most access reviews never see them.

the numbers

all the numbers →

1what AI browser extensions are and what access they hold

AI browser extensions are plugins installed in a browser (Chrome, Edge, Firefox) that add AI-assisted functionality: writing assistance, summarization, tab reading, email drafting, page explanation, and similar tasks.

They are installed by individual users, usually in seconds, from a browser extension store. No IT ticket, no approval, no deployment through endpoint management unless MDM has extension policy controls in place.

The access they request depends on the extension. Common permissions include reading the content of all open tabs, reading text the user types into any form or text field, accessing clipboard contents, and in some configurations reading cookies or interacting with page JavaScript. Some extensions request these permissions broadly ("all websites") rather than limiting themselves to the sites where they are actually useful.

The data that flows through these permissions goes to the extension vendor's infrastructure. The terms of service that govern that flow are the extension vendor's, not your organization's.

2why AI extension sprawl is hard to prevent

Browser extensions are designed to be easy to install. That ease is the same quality that makes them useful and the same quality that makes them a governance problem.

An employee who wants a writing assistant, a grammar checker, or a tab summarizer can install one in under a minute. The install requires no administrator credentials. It does not trigger any IT alert. It does not appear in most endpoint inventories unless MDM extension policies are configured to capture it.

The result is that extension installations accumulate across managed and unmanaged devices without a central record. Employees who leave take their Chrome profiles with them, but the extensions may have already accessed and transmitted substantial amounts of company information while they were active.

3what unmanaged AI browser extensions expose

Data exposure through tab access. An extension with permission to read all open tabs can see everything a user views in the browser: internal documents opened in a web app, sensitive customer data in a CRM, authentication tokens in developer tools, financial reports in a web-based spreadsheet. The extension transmits whatever it reads to the vendor's infrastructure, under the vendor's terms.

No audit trail. A browser extension that accessed company data through an employee's browser leaves no log in your environment. If a data incident later involves that extension, you have no record of what was seen or when.

Authentication token access. Some extensions request permissions that allow access to cookies or JavaScript execution. On pages where authentication tokens are stored, this creates a credential exposure risk. This is not the typical case, but it is a documented attack vector for malicious extensions distributed through legitimate-looking stores.

what it looks like when it goes wrong

Three ways this tends to go, each one taken from a real case.

your company
the gap nobody seesAI extensions can read every page a person opens, including internal tools.company data flowing into AI tools nobody vetted
  1. howEngineers paste code and notes into a free chatbot to get help.
    what leavesconfidential code and meeting noteswho ends up with itan AI vendor on consumer terms
    it happenedSamsung, 2023
  2. howThe AI vendor leaves its own database open.
    what leaveseveryone's chats and the vendor's keyswho ends up with itanyone on the internet
    it happenedDeepSeek exposed database, 2025
  3. howA shared chat link becomes public with one checkbox.
    what leavesclient names and personal detailswho ends up with itanyone using a search engine
    it happenedChatGPT shared links, 2025

4what works

Visibility comes from two places, and they see different things. Directory-side discovery, the IdP's connected-app view, shows only the extensions that hold an OAuth grant to the directory; the full picture of what is installed on devices lives in endpoint management. Organizations running Intune, Jamf, or Google Workspace device management can pull a report of installed extensions, filter it for AI, and read the permissions each one requests. The permission strings that matter are "Read and change all your data on all websites," "Read browsing history," and "Access data on all sites": broad grants that a narrow tool has no reason to hold, and the usual signal that an extension belongs under review or replaced with an alternative that asks for less.

What keeps the inventory from rotting is policy enforced by the browser itself. Chrome Enterprise, Edge enterprise policies, and Google Workspace admin controls all support extension allowlisting, which lets a defined approved list through and stops new installs outside it without blocking extensions wholesale. A clear request path for new approvals keeps the allowlist from becoming a wall people climb over. And because new extensions appear constantly and staff install them independently, the picture only stays true if it is re-pulled on a regular cycle; an inventory from three months ago describes a different browser fleet.

on the record

← all topics

your turn

what you can do about it

  1. this week

    List the browser extensions installed on work browsers, and flag the ones that read page content.

    Your browser management console or a quick staff survey shows them.

  2. go deeper

  3. if you're not sure where to start

    Run the offboarding checklist on one leaver. It covers every account and token to close when someone leaves.

    open the checklist →

the letter · every two weeks

one access gap in your inbox, with the fix.

Every issue takes one way access creeps in, shows a real incident where it went wrong, and ends with one thing to check that week. Written for the people who run the tools, without a security team behind them.

  • 1one mechanism, explained plainly
  • 2one real incident, taken apart
  • 3one step you can take on Monday

get the next issue

One email every two weeks. Your address stays here and is never passed to vendors. One click to unsubscribe. Privacy.