ChatGPT can now read and control your Chrome tabs

For most of the last year, the idea of an AI agent using your browser meant a separate product. OpenAI built one, called it Atlas, and then retired it. On 15 October 2026 the company took the next step in a different direction: instead of asking you to switch browsers, it put the agent inside the one you already use. The upgraded ChatGPT extension for Chrome can now look across your open tabs to answer questions, and in ChatGPT's Work mode it can take control of those tabs to finish tasks for you.

OpenAI announced the change in a series of posts from its official ChatGPT account on X. The Verge reported it the same evening, followed on 16 October by Engadget and 9to5Google. The most useful primary source is OpenAI's own browser extension documentation, which spells out what the extension can touch, how the permission prompts work and what OpenAI stores. This article walks through all of it, with a focus on the part that matters most: what you are actually handing over when you install it.

From Atlas to an extension in the browser you already use

Atlas was OpenAI's attempt at an AI first browser. It launched in October 2025 with ChatGPT at its core. On 9 July 2026 OpenAI said it was deprecating Atlas and moving browser based agentic features into the ChatGPT desktop app, Codex and a Chrome extension. TechCrunch summed up the shift at the time: the browser was a feature, not the destination. According to the OpenAI Help Center, Atlas stopped working on 9 August 2026, and its bookmarks, open tabs and history did not transfer automatically.

The October update is the payoff of that plan. The Chrome Web Store listing, as checked on 16 October, shows version 1.26.1005.11759, updated 15 October 2026, offered by OpenAI, with 7,000,000 users. OpenAI's documentation says the same extension also works in Microsoft Edge, Brave, Opera and Vivaldi. All five support tab mentions and browser control from the desktop app. Side chat, the panel that opens next to the page, is available in all of them except Opera.

What Chat mode does with your tabs

The first change is about reading. When you tag Chrome in the ChatGPT web or desktop app, ChatGPT can now look at the tabs you have open and use them as context. OpenAI's demo showed someone with three sofa listings open asking ChatGPT to compare them. The sample prompts OpenAI published make the intent clear:

• Compare the three couches I have open by dimensions, material, delivery time and return policy.
• Explain this article in plain English, including the main argument and the evidence behind it.
• Compare these pricing pages against our requirements and say which plan covers everything.

The documentation adds a few smaller tools. You can highlight text on a page and pull just that selection into a chat. You can right click a page and choose Ask ChatGPT. On YouTube, side chat can use a video's timestamped transcript when captions are available. Chats you start in side chat show up in the main ChatGPT app, and recent chats from the app can be reopened in the side panel, so the conversation follows you between the browser and the desktop.

OpenAI also says the in browser ChatGPT now carries your entire conversation history, so the side panel is no longer a thin helper with a blank memory. Per the documentation, it follows your Memories setting: if Memories is on, relevant saved memories can be used while ChatGPT works in your browser. If Memories is off, browser control does not use them.

What Work mode does with your tabs

The second change is about acting. ChatGPT Work is OpenAI's agent for longer, multi step tasks. With the upgraded extension, Work can control Chrome tabs to do things like sending messages, filling out forms and downloading files. It can also organize and group the pile of tabs you have open. OpenAI's examples here are more ambitious:

• Use the details in an event brief to fill out a registration form, and stop before submitting.
• Put the vendors in these tabs into a comparison sheet with pricing, key features and source links.
• Compare the contractor services across my open tabs, then draft an email to the best fit.
• Fill out the form I have open ahead of a first dentist visit, and add the appointment to my work and personal calendars.

The store listing describes the same capabilities in broader terms: research across live and authenticated websites, creating and editing documents, spreadsheets and slides, data entry into record systems such as timesheets, inventory tools and content management systems, filing expenses, downloading files and images in bulk, testing websites, and cleaning up tabs. It says ChatGPT can do this work in the background without taking over your active browsing session.

The documentation is explicit that this works on sites where you are already signed in. It names LinkedIn, Salesforce, Gmail and internal company tools as examples. That is the whole point of the feature, and also the reason to think carefully before turning it on. A built in agent browser, which OpenAI still offers inside ChatGPT through an @Browser mention, keeps that work away from your regular profile. The Chrome extension is the opposite: it works with your real sessions.

The permission prompt: four choices per website

OpenAI has put a layer of confirmation on top of what the extension can technically do. By default ChatGPT asks before it interacts with each new website, and the prompt is based on the host, such as example.com. You get four answers:

• Decline: ChatGPT does not use the site.
• Allow once: one use, then it asks again next time.
• Allow for this site: the domain goes on an allowlist, and ChatGPT stops asking there.
• Allow for all sites: ChatGPT no longer asks for confirmation on any website.

OpenAI's own wording on the last option is blunt: only choose it if you trust ChatGPT to use any website open in the browser. The allowlist and a matching blocklist live in the desktop app under Settings, Computer Use, Manage, and they are shared across all supported browsers.

Browser history gets special treatment. The documentation warns that history can contain sensitive telemetry, internal URLs and search terms from all your signed in devices. ChatGPT asks each time it wants to use history, scopes that access to the request, and there is no always allow option for it. The store listing adds that ChatGPT will always ask before referencing browser history and before downloading or uploading files. File uploads also require you to switch on Chrome's own "Allow access to file URLs" setting for the extension.

What Chrome is asked to allow at install

Under the ChatGPT prompts sits the extension itself, and its Chrome permissions are about as wide as an extension can ask for. OpenAI's documentation lists what the install prompt may include: access to the page debugger, reading and changing all your data on all websites, reading and changing your browsing history on all signed in devices, displaying notifications, reading and changing bookmarks, managing downloads, communicating with cooperating native applications, and viewing and managing tab groups.

OpenAI's position is that these permissions are what make browser workflows possible, and that ChatGPT still applies its own confirmations, allowlists and blocklists before it uses a site or your history. That is a reasonable design, but it is worth being precise about it. The Chrome permission is the hard ceiling. The ChatGPT prompts are a software policy that sits under that ceiling and that you can loosen with a single click on Allow for all sites.

The store listing's privacy section is also worth reading. OpenAI declares that the extension handles personally identifiable information, health information, financial and payment information, authentication information, personal communications, location, web history, user activity and website content. It states that this data is not sold to third parties outside approved use cases and is not used for creditworthiness or lending decisions. At the time of checking, the listing had an average rating of 2.2 out of 5 from 922 ratings.

What OpenAI stores, and what it does not

The documentation draws a line that matters for privacy. OpenAI says it does not keep a separate, complete record of your browser actions from the extension. Browser activity is stored only when it becomes part of the ChatGPT context, for example text ChatGPT reads from a page, screenshots, tool calls, summaries or messages in the chat. Your normal ChatGPT data controls then apply to that content.

In practice, that means the record of a task is the chat itself. If ChatGPT reads your banking page to fill in an expense report, whatever it read can end up in that conversation. OpenAI's advice is to avoid sending secrets or highly sensitive data through browser tasks unless they are required and you are present to review each prompt.

Why it matters

There are three reasons this update is bigger than a feature bump.

First, scale. Atlas asked people to adopt a new browser. The extension meets 7,000,000 existing users where they already are, and OpenAI says it is available on all ChatGPT plans. Agentic browsing just went from a niche product to a default option inside the most popular browser in the world.

Second, competition. The Verge compared Work's tab control to Google's Gemini powered auto browse feature, and 9to5Google pointed out that the new side panel looks a lot like Chrome's built in Gemini. TechCrunch called the original extension a direct competitor to the Gemini side panel back in July. OpenAI is now competing with Google inside Google's own browser.

Third, security. A browser agent with access to signed in sessions is exactly the kind of system that prompt injection was invented to abuse. OpenAI's own documentation tells users to treat webpage content, selected text and video transcripts as untrusted, and warns that malicious or misleading page content can increase the risk that ChatGPT copies data somewhere unintended. That is an honest warning. It is also a reminder that the safety of the setup depends heavily on the permission choices people make on day one. Regulators are already paying attention to agent incidents: the US Federal Trade Commission this week sent civil investigative demands to OpenAI and Anthropic about AI agent safety and security, which we covered here.

Practical tips if you turn it on

• Keep the default. Let ChatGPT ask per site, and use Allow once for anything that touches money, health or work accounts.
• Do not click Allow for all sites in a profile where you are signed in to email, banking or admin consoles.
• Consider a separate Chrome profile for agent tasks. OpenAI's docs say the extension works in the profile where you installed it, so a clean profile limits what it can reach.
• Use the blocklist for domains that should never be touched, such as your password manager's web vault or internal admin tools.
• Ask the agent to stop before submitting. OpenAI's own sample prompts do this, and it keeps a human click on the final action.
• Remember that Memories carry over. If you do not want saved memories used while ChatGPT browses, turn Memories off before running a task.

Dany's take

I like the direction and I am nervous about the defaults people will choose. Putting the agent into Chrome instead of a separate browser is the right product call, because almost nobody wants to switch browsers for one feature. Tab context in Chat is genuinely useful and fairly low risk, since it only reads what you already opened. Work mode is a different animal. An agent that can fill forms and send messages inside your logged in Gmail or LinkedIn is powerful, and the only thing standing between a helpful task and a bad one is a permission prompt that many users will click through. My rule for now: Chat for reading, Work only in a clean profile, and never Allow for all sites.

The bottom line

OpenAI's ChatGPT extension for Chrome can now read your open tabs in Chat and control them in Work, using your real signed in sessions. It ships with very broad Chrome permissions and a per site confirmation layer on top. The feature is useful, it is available on every plan, and it puts OpenAI head to head with Gemini inside Chrome. How safe it is will depend less on the model and more on whether users keep the guard rails switched on.

Sources: OpenAI browser extension documentation, Chrome Web Store listing, The Verge, Engadget, 9to5Google, OpenAI Help Center on the Atlas retirement, TechCrunch on the Atlas shutdown.