Fourteen hours ago, OpenAI quietly gave ChatGPT the power to read and act inside your open Chrome tabs. Not a new browser. Not a new app. The extension you already have, upgraded to version 1.26.1005.11759, with a new Work mode that can send messages, fill forms, download files, and group tabs on its own. The change is small on the surface and enormous underneath, and it arrived without a keynote. Most users will not notice the version bump. They will notice that their browser can now act on their behalf, and that is the kind of shift that rewrites how you think about a tool you have used for years.
What actually changed on October 15
The ChatGPT Chrome extension already had about seven million users and 915 ratings before this update. What changed is how much of ChatGPT now lives inside the browser, and who gets it. In Chat mode, ChatGPT pulls data from several tabs at once. In Work mode, it controls tabs through Computer Use. GPT-6 Astra, which OpenAI calls the best Computer Use model on the market right now, powers both. It handles complex multi-step workflows that earlier models fumbled, and it does so without asking you to leave the page you are reading.
The extension now works with all ChatGPT plans. The setup lives in the ChatGPT desktop app: open Settings, then Computer Use, select the browser, and install from the store page. The sidebar puts full ChatGPT, with your complete conversation history, next to whatever page you are viewing. That sidebar is the part that matters most. It means the assistant is always one click away, on the page you are already reading, instead of in a separate tab you have to context-switch into.
The two modes and what each actually does
In Chat mode, the assistant looks across your open tabs and gathers information. Open three sofa listings in three shops, and it can compare them by size, material, shipping time, and return rules without you copying anything. Summarize an article in plain language. Match price pages from different sellers. The promise is the boring, useful work that eats an afternoon, and it is the kind of task that is tedious for a human and trivial for a model that can read three tabs at once.
In Work mode, the agent goes further. It controls the tabs themselves: sending messages, filling out forms, downloading files, grouping a messy browser into tab groups. The extension's store listing says it works in the background, without taking over your active browsing session. OpenAI says it can read and act on sites where you are already signed in, including LinkedIn, Salesforce, Gmail, and internal tools. That last point is the one that should make you stop and think. The agent is not just reading public pages. It is operating inside your authenticated sessions, which is where the real work, and the real risk, actually lives.
Why this is a bigger deal than it looks
The browser is the last place where most knowledge work actually happens. Every other agent product has asked you to leave your browser and work in their interface. This one comes to where you already are. That inversion is the whole story, and it is the reason this release matters more than its quiet launch suggests.
There is also a continuity angle. OpenAI is sunsetting the standalone Atlas browser and migrating users to this extension. The lesson they learned from Atlas, and from users who helped shape how agentic tools make browser-based work useful, is that people do not want another browser. They want their existing browser to do more. That is a quiet concession from a company that spent a year building its own, and it is the kind of strategic retreat that usually signals a category has shifted underneath the players.
The permission tradeoff nobody mentions
Here is the part that should make you pause. The Chrome Web Store listing asks for the ability to "read and change all your data on all websites" and to "read and change browsing history on all signed-in devices." That is not a narrow permission. It is the broadest one Chrome offers, and it is the price of letting an agent act inside your accounts.
The extension works in the background, which is convenient, but it also means the agent can touch signed-in sessions while you are doing something else. OpenAI's guide says it can read or act on sites where you are already signed in, including internal tools. That is a feature for productivity and a liability for security teams. If you are weighing this, the right question is not whether the feature is useful. It is whether your organization's risk posture allows an agent with full read-write access to every tab, and whether your users understand what they are consenting to when they click install. For a solo developer that tradeoff is easy to accept. For a company with regulated data, it is a procurement conversation, not a settings toggle.
What it gets right, and where it still frays
The strengths are real. Cross-tab comparison, summarization, and form filling are the tedious tasks that agents should own. Running in the background without hijacking the active session is the right call. Working in Edge, Brave, Opera, and Vivaldi on desktop is broader than the Chrome-only branding suggests, and that breadth matters for teams that do not standardize on a single browser.
The rough edges are also real. The extension is not new. It is version 1.26.1005.11759 with 915 store ratings. What changed is the depth, not the presence. Availability of each mode can depend on rollout and your workspace settings, so the experience is not uniform. The guide acknowledges that availability "can depend on rollout and your workspace settings." That hedging is honest, but it means the headline feature may not be yours yet, and a feature you cannot reliably reach is not a feature you can plan around. The real test will be whether Work mode can reliably complete multi-step tasks without supervision, and whether users trust it enough to leave running in the background while they do something else.
What this means for the next wave of browser agents
The browser is becoming the agent's home, not its destination. Every tool that asked you to leave the browser is now competing with one that lives where you already work. The open question is not whether agents can act in browsers. It is whether users will grant the permissions required to make that action useful, and whether the security teams will let them.
For tool builders, the lesson is direct. Meet users where they already work, and make the permission tradeoff explicit rather than buried in a store listing. The teams that win the browser agent category will be the ones that earn trust before they ask for access, and that treat the browser as the product surface instead of a launchpad to somewhere else.
Which browser agent are you letting touch your tabs right now? Base the choice on the hardware and the permissions you are comfortable with, not on the size of the model card.
Top comments (0)