OpenAI's ChatGPT Chrome extension can now read every tab you have open and, in a new Work mode, act inside them. It can send messages, fill out forms and download files. For roughly 7 million Chrome users, that turns a helper that answered questions into a browser agent.
OpenAI announced the update on October 15, 2026. According to Unite.ai, the extension can look across open browser tabs, take control of them to complete tasks, and organize and group tabs. OpenAI framed the release on its official ChatGPT account on X as "bringing the power of computer use to anyone who uses Chrome."
The Chrome Web Store listing shows version 1.26.1005.11759, updated October 15, with 7,000,000 users. That count comes from the store page, not from an OpenAI usage disclosure. It shows how many people installed the extension, and it says nothing about how many use it.
The update also absorbs capabilities from Atlas, OpenAI's standalone browser, which the company shut down earlier this year. Atlas is gone, and the Chrome extension now serves as the main route for ChatGPT into a user's browser, according to ExplainX.
Who is affected and what changes for them
The people most affected are US Chrome users who sign in to email, social, shopping, HR or CRM tools every day. Small business owners who already use ChatGPT for admin work also stand to gain the most. For them, data entry and tab cleanup are routine chores that eat hours, and an agent that handles them looks attractive.
Chat mode and Work mode, explained
The extension now runs two modes, and the difference determines how much access you hand over. Chat reads across open tabs to gather information for an answer. Work goes further and controls the tabs, sending messages, filling out forms, downloading files and grouping tabs.
Work is built for longer, multistep jobs. OpenAI's examples include registration forms, calendar entries, data entry into business systems and tab cleanup, as IBTimes Australia reported. Those chores drag on because they repeat across many pages.
The extension can act on sites where you are already logged in, so it can reach services like Gmail, LinkedIn or Salesforce, provided the permissions are in place. That is the feature's appeal and its exposure. An agent running in your signed-in session does not need a separate login, which makes the permission settings matter more than any single button.
OpenAI says the extension works in the background and does not take over the active browsing session, according to Hans India. That claim sits uneasily beside the announcement's own language about taking control of tabs. Test the behavior on a low-stakes site before you trust it with a live workflow.
Which plans get Work mode
Plan access is the least settled detail. IBTimes Australia reports OpenAI saying the feature covers all ChatGPT plans. A separate Powerdrill guide says Work was rolling out to paid plans, excluding Free and Go, citing OpenAI's July 9 notes.
A rollout that expanded after July could reconcile the two accounts, but the sources reviewed do not confirm that. Availability outside the United States was also not verified. If you are on a free or Go plan, check your ChatGPT settings before planning around Work mode.
How to set up the ChatGPT Chrome extension
Setup runs through the ChatGPT desktop app. The steps reported by IBTimes Australia and Powerdrill are short:
- Install the extension in Chrome.
- Enable computer use in the ChatGPT settings.
- Select the browser the agent should use.
- Open a Work chat and pick that browser with an @-mention.
Browser support reaches beyond Chrome. Hans India reports the update also works in Edge, Brave, Opera and Vivaldi, though side chat is not available in Opera. Confirm the feature behaves as expected in your browser before you build a routine around it.
Security risks and permission settings to check
Permission settings deserve more attention than the feature list. Setup coverage from WindowsForum says OpenAI recommends verifying the blocklist during setup, which makes it the first thing to review.
ExplainX flags security and permission risks around Work mode, but the sources reviewed did not include independent security testing, including prompt-injection risk. No specific exploit has been verified for this release. It is also unclear whether Work asks for confirmation before each send on an allowed site. Downloading files adds a second risk, because an agent that can fetch files from a page can fetch a malicious one too.
Practical steps to take now:
- Review the list of sites the extension can act on, and remove any you do not use.
- Verify the blocklist during setup, as OpenAI recommends.
- Keep banking accounts and other financial sites off the allowed list.
- Confirm your plan includes Work before you rely on it.
- Trial Work on a low-stakes form before letting it touch HR, CRM or email tools.
The last two steps are general practice rather than part of OpenAI's own guidance, so treat them as a starting point.
Why banks blocking agents makes this a bigger story
The Chrome extension arrives as banks and websites push outside agents away. According to AI Weekly's roundup of The Verge's reporting, JPMorgan Chase and Charles Schwab bar third-party agents from account access, and Delta has no deals with third-party agents. USA 24 has covered the banking side in big US banks are blocking AI agents from moving customer money.
The consumer browser changes the terms of that fight. A bank can refuse an outside agent at its own login page. It has far less say over an agent that runs on your machine, inside a session you opened yourself. The question shifts from "can an agent reach my account" to "what permissions did I grant my browser."
India's cyber agency has raised a related concern. USA 24 has reported that India's CERT-In says AI agents read live business records unnoticed, an exposure that matters just as much inside a company as on a personal browser.
The Atlas shutdown explains the timing. OpenAI gave up a standalone browser, and the extension now carries that ambition inside Chrome, where users already spend their day. The bet is convenience over a new app, with the browser's existing sessions doing the heavy lifting.
What to watch next
Three open questions will decide how this lands. First, whether Work reaches every plan or stays limited to paid tiers. Second, whether independent researchers publish testing on prompt injection and on how data from signed-in sites is handled. Third, how many people actively use the feature, since the 7 million figure only measures installs.
Watch also whether Work adds a confirmation step before sends and form submissions, and whether banks and sites that already block agents extend those blocks to Chrome-based tools. The coming weeks should show whether OpenAI treats the extension as a productivity feature or as a managed channel into customer accounts.
