ChatGPT's Chrome extension can now read your open tabs and act in them

What OpenAI changed on 15 October

OpenAI's ChatGPT account announced the change in a thread on X at about 20:02 UTC on 15 October 2026, saying the company is "bringing the power of computer use to anyone who uses Chrome." The thread describes two upgrades. First, ChatGPT can use the tabs you already have open, either to gather information or to get tasks done. Second, ChatGPT running inside a Chrome tab "got a major upgrade": the full app, with your entire conversation history, now sits beside the site you are on.

A follow-up post links to the extension and says it is available for all ChatGPT plans. The Chrome Web Store listing shows version 1.26.1005.11759, last updated on 15 October 2026, offered by OpenAI, with about seven million users and an average rating of 2.4 from 915 ratings at the time we checked. The listing now leads with "Control your browser with ChatGPT."

OpenAI did not publish a blog post or press release for this update. The detailed behaviour is documented on the "Browser extension" page of its ChatGPT Learn documentation, which this article relies on alongside the X thread and the store listing. The Verge reported the launch the same evening and noted that OpenAI's separate ChatGPT Atlas browser is no longer offered; we did not independently confirm Atlas's status.

Chat reads your tabs, Work acts in them

The update splits browser help along ChatGPT's two modes. In Chat, the everyday conversational mode, OpenAI says ChatGPT "can look across your open tabs and gather info." That is read-only context: compare three product pages, summarise an article, or check a pricing page against a list of requirements without copying anything by hand.

In Work, the agent mode OpenAI introduced for longer tasks, ChatGPT can "control your Chrome tabs" to send messages, fill out forms and download files, and it can organize and group a pile of open tabs. The documentation says ChatGPT can read or act on sites where you are already signed in, naming LinkedIn, Salesforce, Gmail and internal tools as examples. That is the practical difference from a cloud agent: it uses your own logged-in browser profile rather than a fresh, anonymous browser in a data centre.

OpenAI's example prompts show where it expects the line between the two modes to fall. Several are pure comparison ("Compare the three couches I have open by dimensions, material, delivery time, and return policy"). Others are actions: fill out a registration form from an event brief "and stop before submitting," put vendors from several tabs into a comparison sheet with source links, or fill in a new-patient form and add the appointment to two calendars. The phrase "stop before submitting" is a hint that OpenAI expects users to keep the final click for themselves on consequential forms.

The Chrome Web Store description adds that ChatGPT works in the browser "alongside you, in the background, without ever taking over your active browsing session," and that it will always ask permission before referencing browser history or downloading and uploading files.

Side chat puts the full app next to the page

The second half of the update is side chat. According to the documentation, you open ChatGPT from the browser toolbar or Extensions menu, or press Cmd+Shift+. on macOS, and a panel appears next to the page. You can ask about the page, highlight a passage and bring just that selection into the chat, or right-click and choose Ask ChatGPT. On a YouTube video with captions, ChatGPT can use the timestamped transcript to summarise or answer questions.

The panel stays attached to the tab where you opened it. Chats started there appear in the main ChatGPT app, and recent chats from the app can be reopened in the side panel, so a task can move between the two. OpenAI's X post frames this as "the full power of ChatGPT" in the browser, with Chat or Work available side by side while you stay on a site.

Side chat is supported in Chrome, Edge, Brave and Vivaldi. Opera supports tab mentions and browser control from the desktop app but not the side panel. The Chrome Web Store listing also lists Edge, Brave and Opera as compatible Chromium-based browsers.

How to set it up

Getting the full feature set takes two pieces. You install the ChatGPT extension from the Chrome Web Store, and for tab control you also connect it to the ChatGPT desktop app. The documentation walks through Settings, then Computer Use, in the desktop app: pick the browser, select Install to open its extension store page, accept the permission prompts, and confirm the browser then shows Manage.

After that, you start a ChatGPT Work or Codex chat and choose the browser with an @-mention, for example "@Edge open Salesforce and update the account from these call notes." You can also mention a specific open tab to give ChatGPT context from that page. The extension has to be installed in the browser profile you are actually using, and a per-browser toggle controls whether it appears in the @-mention menu.

OpenAI notes that availability can depend on rollout and workspace settings, so business and enterprise users may find the feature controlled by an administrator. ChatGPT also keeps a separate built-in browser, reached with @Browser, that works without your regular profile. The docs say ChatGPT can switch between plugins, your own browser and the built-in browser depending on what a task needs.

The permissions you are handing over

Acting inside a signed-in browser requires broad access, and OpenAI's documentation is unusually direct about it. Chrome's install prompt may ask to access the page debugger, read and change all your data on all websites, read and change browsing history on all signed-in devices, manage downloads and bookmarks, communicate with native applications and manage tab groups.

On top of those browser permissions, ChatGPT applies its own checks. By default it asks before interacting with each new website, keyed to the host name. Users can allow once, allow for that site, allow for all sites, or decline, and can manage an allowlist and blocklist in the desktop app's Computer Use settings. OpenAI labels "Allow for all sites" as elevated risk, since ChatGPT then stops asking before it uses websites.

Browser history gets a stricter rule. OpenAI warns that history can contain internal URLs, search terms and activity from other signed-in devices, and that malicious page content could trick ChatGPT into copying that data somewhere unintended. ChatGPT asks each time it wants history, scopes it to the request, and offers no always-allow option.

OpenAI says it does not keep a separate, complete log of browser actions; it stores browsing activity only when it becomes part of the chat, such as page text, screenshots, tool calls or summaries, and normal ChatGPT data controls apply. The docs repeatedly tell users to treat page content as untrusted and to avoid sending secrets through browser tasks unless they are present to review each prompt. That warning echoes the prompt-injection risks covered in our explainer on computer-use agents.

Why this matters for ChatGPT Work

The update gives ChatGPT Work a much larger surface to act on. OpenAI has been pushing agentic use hard: its own report on European business customers said Codex and ChatGPT Work produced 55 percent of EU business output tokens in September. Until now, many everyday office tasks still lived in web apps that an agent could only reach through a separate browser or a dedicated plugin.

Plugging into the browser people already use, with their existing sessions, removes a setup step for expense systems, CRMs and internal tools that have no plugin. It also puts ChatGPT in more direct competition with Google, whose Gemini-powered auto browse works inside Chrome, and with the workplace agents Google Cloud described at its Gemini at Work keynote. OpenAI's own research has shown how hard long browser workflows remain: its Ironclad evaluation tested GPT-6 Astra on multi-step contract tasks inside a real enterprise app.

Open questions and limits

Several practical details are still unclear. OpenAI says the extension works on all plans, but neither the X thread nor the documentation spells out whether free users get Work-mode tab control or only Chat-mode tab reading, or whether usage limits differ by plan. It also does not say which model drives browser actions or publish any success rate for tasks like form filling.

The documentation does not say whether Work shows the exact text of a message before it sends it on your behalf, beyond the general per-site permission prompts. Until that is clearer, the safest pattern is the one OpenAI's own example uses: ask ChatGPT to prepare the form or draft and stop before the final submit or send.

The store listing's 2.4-star average is worth noting but hard to interpret, since ratings accumulate across older versions of the extension. The more useful test is personal: start with read-only comparisons in Chat, keep per-site prompts on, block sensitive domains such as banking and HR systems, and only then hand Work a routine task on a site where a mistake is easy to undo.