Artificial Intelligence (AI)

The ChatGPT Chrome Extension Now Acts on Your Tabs: What IT Should Allow and Block

An editorial photograph of a stone office entrance with a heavy door propped open to a sunlit courtyard and five concrete shapes resting near the threshold, a visual for the ChatGPT Chrome extension, which can now act in open tabs, and the site access an IT team decides to allow

The ChatGPT Chrome extension can now read your open Chrome tabs and act on them: fill in forms, send messages, download files and group tabs. OpenAI announced the update on October 15, 2026, and the Chrome Web Store listing shows a new version the same day.

According to PiunikaWeb, this is where OpenAI moved the tab features from its retired Atlas browser. It now runs inside the browser your staff already use. This post covers what the extension does, how its permission prompts work, which controls OpenAI gives admins, and what an IT lead at a nonprofit or association should set before staff turn it on. Details are as of October 19, 2026.

The short version: ChatGPT asks before it uses each new website, and users can allow once, allow a site, or allow all sites. Enterprise and Edu admins can set site rules that users cannot override. Everyone else should lean on Chrome’s own extension policy, which can block the extension from your donor database, payment processor and payroll sites. One gap matters: when ChatGPT drives a browser through Computer Use instead of the extension, OpenAI says the browser site policies do not apply.

What the ChatGPT Chrome extension does now

According to 9to5Google and PPC Land, OpenAI’s announcement on X described two modes. In Chat, ChatGPT reads across your open tabs and gathers information, such as comparing pricing pages. In Work, it controls your tabs. 9to5Google quotes OpenAI: "ChatGPT can control your Chrome tabs to get things, like sending messages, filling out forms, or downloading files."

The update also brings a side panel. OpenAI’s browser extension documentation calls it side chat. It stays with the tab where you opened it, and chats started there show up in the ChatGPT app. You can mention an open tab, highlight text to bring it into the chat, or right-click a page and choose "Ask ChatGPT."

The Chrome Web Store listing, updated October 15, describes the extension as "Control your browser with ChatGPT." It says the extension works "in the background, without ever taking over your active browsing session." As of October 19, the listing showed version 1.26.1005.11759 and 7,000,000 users.

OpenAI’s documentation says browser tasks start in a ChatGPT Work or Codex chat. The same extension works in Edge, Brave, Opera and Vivaldi. Side chat is not available in Opera.

Chat reads, Work acts

The split between the two modes is the first thing to explain to staff who install the ChatGPT Chrome extension.

Mode What it does with tabs Risk to plan for
Chat Reads open tabs and answers questions about them Page content leaves the browser and becomes chat context
Work Clicks, types, fills forms, sends messages, downloads files, groups tabs Actions run under the user’s signed-in accounts
Codex Can start browser tasks from a Codex chat, per OpenAI’s docs Same site prompts as Work; mainly a developer concern

Reading is lower risk than acting, but it is not zero risk. Anything ChatGPT reads becomes part of the conversation. OpenAI’s documentation says it does not keep a separate full record of browser actions. It stores browser activity only when it becomes chat context, such as page text, screenshots or tool calls.

Acting is where the real exposure sits. The ChatGPT Chrome extension works in tabs where the user is already signed in. OpenAI’s own examples include LinkedIn, Salesforce and Gmail. For a nonprofit, that list could include the association management system, the donor CRM, the email platform and the bank.

How the ChatGPT Chrome extension asks for permission

OpenAI’s documentation describes three layers of permission.

Chrome’s install prompt. Installing the extension shows Chrome’s standard permission list. OpenAI says it may include reading and changing data on all websites, page debugger access, browsing history, notifications, bookmarks, downloads, native app communication and tab groups. This is broad, and it is the same kind of access other agent extensions request.

Per-site prompts. ChatGPT asks before it uses each new website, matched by host, such as example.com. The choices are:

  • Allow once for this task only.
  • Allow for this site, which adds the host to an allowlist.
  • Allow for all sites, which OpenAI marks as elevated risk. ChatGPT stops asking.
  • Decline.

Users can edit the allowlist and blocklist under Settings, Computer Use, Manage. All supported browsers share the same lists.

History. Browser history is also marked elevated risk. ChatGPT asks each time it wants history, limits access to the request, and offers no always-allow option. PiunikaWeb reports that OpenAI says it will also ask before managing downloads.

OpenAI’s documentation tells users to treat page content as untrusted. A page can contain hidden instructions. We covered why that matters in our post on agentic browser security.

Plans and availability

OpenAI’s pages do not give a plan table for the ChatGPT Chrome extension. Here is what we could confirm and what we could not.

According to PPC Land, OpenAI’s announcement said the extension is "Available for all ChatGPT plans." The extension is free to install. PiunikaWeb reports that Work mode needs at least a Plus plan. We could not confirm that on an OpenAI page.

OpenAI’s documentation says browser availability "can depend on rollout and your workspace settings." If you run a ChatGPT Business, Enterprise or Edu workspace, check your workspace settings before you assume staff can or cannot use it.

For background on how Chat, Work and Codex fit together, see our explainer on the new ChatGPT: Chat, Work and Codex and our earlier post on what ChatGPT Work is.

The admin controls OpenAI provides

OpenAI’s help article on managing browser and computer use in an Enterprise workspace covers Browser Use through the Chrome extension. Admins with the right Codex admin permission set it under Policies & Configurations.

The main controls:

  • Browser Use site rules. Set a default of Allow or Block, then add exceptions by origin, such as `https://example.com`. If rules conflict, Block wins.
  • Per-site permissions. Site access, uploads, downloads, and debug access can each be set per site.
  • Approvals. Admins can turn on auto-review, set how long an approval lasts (one turn or one thread), and control saved “Always allow” approvals.
  • Groups. Policies apply to the whole workspace or to selected groups.

OpenAI’s ChatGPT Work local security guide adds that "A user approval cannot override an administrator’s denial." It also lists managed settings in requirements.toml. Setting features.browser_use_external = false turns off agent browser control through extensions. Setting features.computer_use = false turns off Computer Use.

The Computer Use gap

One line in the local security guide matters more than the rest. When Computer Use operates a browser directly, OpenAI says "Browser Use’s site and capability policies do not apply to this path."

In plain terms, your site rules govern the extension. They do not govern ChatGPT moving the mouse and keyboard in a browser window through desktop Computer Use. Computer Use has its own controls: admins can limit which apps it may operate and whether app approvals can be saved.

If you block your payment processor in Browser Use site rules but leave Computer Use allowed for Chrome, the block has a way around it. Either restrict Computer Use from operating browsers, or turn Computer Use off for groups that work in sensitive systems.

Chrome policy: the control every IT lead already has

Many small nonprofits do not have an Enterprise workspace. OpenAI’s admin article is written for Enterprise workspaces and does not say whether Business admins get the same page. Chrome’s own policy works either way, as long as you manage Chrome.

Google’s ExtensionSettings policy lets you set rules for one extension by its ID. The ChatGPT Chrome extension ID, from its Chrome Web Store address, is hehggadaopoacecdllhhajmbjkdcmajg.

Three settings do most of the work:

  • `installation_mode`. `blocked` stops installs. `removed` blocks it and removes existing copies. `force_installed` pushes it to everyone and stops removal.
  • `runtime_blocked_hosts`. Stops the extension from reading or changing listed sites. Up to 100 entries.
  • `runtime_allowed_hosts`. Lets it work on listed sites even if they match a block. Up to 100 entries.

Host patterns look like *://*.example.com and cannot include a path. Chrome enforces these rules no matter what a user clicks in ChatGPT’s own prompt.

A starting configuration for a nonprofit

This is a cautious starting point. Adjust it after a pilot.

  1. Decide who gets it. Allow the ChatGPT Chrome extension for a pilot group. Block it for everyone else with `installation_mode`.
  2. Block sensitive sites at the Chrome level. Put these in `runtime_blocked_hosts`: your AMS, donor CRM, payment processor, payroll and HR system, online banking, accounting system and password manager.
  3. Set a Block default in Browser Use site rules if you have an Enterprise or Edu workspace, then allow low-risk sites.
  4. Turn off saved “Always allow” approvals and keep approvals to one turn.
  5. Block downloads and uploads by default in per-site permissions.
  6. Close the Computer Use gap by limiting which apps it can operate.
  7. Tell staff never to pick “Allow for all sites.” Chrome policy cannot see that choice; it can only stop the extension from reaching blocked hosts.
  8. Suggest a separate Chrome profile for agent work, with only the accounts that task needs.

Google’s 100-entry limit is enough for most small organizations that pilot the ChatGPT Chrome extension. List each system’s login and app domains.

ChatGPT Chrome extension vs Claude in Chrome and Gemini in Chrome

OpenAI moved its browser agent into Chrome after its decision to retire the ChatGPT Atlas browser. Anthropic and Google already work this way. Anthropic’s Claude in Chrome is an extension with per-site prompts, which we tested in our hands-on look at Claude for Chrome. Google builds Gemini into Chrome and calls its agent feature auto browse. Our 2026 agentic browser landscape covers the wider field.

Product User site prompts Admin site controls Admin default
ChatGPT Chrome extension Allow once, this site, all sites, or decline Browser Use site rules (Enterprise workspace) Not stated by OpenAI
Claude in Chrome Allow this action, always allow this site, or decline Allowlist and blocklist in Organization settings (Team, Enterprise) On for Team and Enterprise
Gemini in Chrome auto browse Managed in Chrome GeminiActOnWebAllowedForURLs and GeminiActOnWebBlockedForURLs policies Off for managed users

Sources: Anthropic’s Claude in Chrome admin controls and Google’s Gemini in Chrome admin guide. Google’s is the only one of the three that ships with agent actions off by default for managed users.

Frequently Asked Questions

What changed in the ChatGPT Chrome extension in October 2026?

OpenAI announced on October 15, 2026 that ChatGPT can read open Chrome tabs and act on them. It can fill forms, send messages, download files and group tabs. The update also added a side panel with your ChatGPT history.

What is the difference between Chat and Work in the extension?

In Chat, ChatGPT reads your open tabs and answers questions about them. In Work, it controls tabs and takes actions, such as filling in a form or sending a message.

Does ChatGPT ask before it uses a website?

Yes. It asks before each new website, by host. Users can allow once, allow for this site, allow for all sites, or decline. OpenAI marks “Allow for all sites” as elevated risk.

Which ChatGPT plans can use it?

According to PPC Land, OpenAI’s announcement says the extension is available on all plans. PiunikaWeb reports Work mode needs Plus or higher. OpenAI’s own pages say availability can depend on rollout and workspace settings.

Can admins block specific sites?

Yes, in two places. Enterprise workspace admins can set Browser Use site rules in ChatGPT. Any organization that manages Chrome can use `runtime_blocked_hosts` in the ExtensionSettings policy.

What is the ChatGPT extension ID for Chrome policy?

`hehggadaopoacecdllhhajmbjkdcmajg`, taken from its Chrome Web Store address. Use it as the key in the ExtensionSettings policy.

Do browser site rules cover Computer Use?

No. OpenAI’s local security guide says Browser Use site and capability policies do not apply when Computer Use operates a browser. Use Computer Use app rules to limit it.

Does OpenAI keep a log of what ChatGPT does in my browser?

OpenAI says it does not keep a separate full record of browser actions. Page text, screenshots and tool calls are stored when they become part of the chat, under your normal ChatGPT data controls.

Does it work in browsers other than Chrome?

Yes. OpenAI’s documentation lists Edge, Brave, Opera and Vivaldi. Opera does not support side chat, so tasks there start in the ChatGPT desktop app.

Digital Matters

Artificial Intelligence (AI) Desk