ChatGPT's Chrome Extension Can Now Control Your Browser Tabs

Install OpenAI's Chrome extension this week and you're not just adding a sidebar chatbot to your browser. You're handing an AI agent permission to read every open tab, click buttons, fill in forms, and submit them, without you touching the keyboard.

From reading your tabs to acting on them

OpenAI rolled out an update to the ChatGPT Chrome extension that splits it into two distinct modes. The first, called Chat, does what you'd expect from an AI assistant: it looks across whatever tabs you have open and answers questions about them, say, comparing prices or specs from three different shopping tabs without you copy-pasting anything.

The second mode, Work, goes further. This is what's known as an agentic feature, meaning the AI doesn't just respond to your question, it takes actions on its own, in sequence, to get a task done. In Work mode, ChatGPT can fill out web forms, send messages, download files, and organize your open tabs into groups. Point it at a task and it drives the browser itself.

It's a sign of where the big AI labs are pushing hardest right now: not smarter chatbots, but agents that act inside the tools you already use. OpenAI tried a standalone AI browser earlier this year; this extension is a quieter, lower-friction way to get the same capability in front of people who already have Chrome open.

The permission screen most people will click past

To work across your tabs, the extension needs broad access, and Chrome says so plainly before you install it. Reporting on the rollout has flagged the exact wording users see:

"Read and change all your data on all websites"

That's the literal permission Chrome's own installer shows for this extension. OpenAI has added a layer of friction on top of it, the extension reportedly asks for your go-ahead before it interacts with a new website for the first time, rather than acting silently across your whole browsing history. But the underlying grant is still sweeping. Practically, that means the agent could, in principle, see or touch whatever sits in your other open tabs:

• Webmail and messaging tabs it might be asked to act in
• Shopping or banking sessions sitting open alongside it
• Any form field, including ones holding personal or financial details

None of this means the extension is misusing that access today. It means the access itself is now sitting in millions of browsers, and how it's used depends entirely on how carefully the product is built and how carefully each person configures it.

What this means if you're browsing from India

This matters more in India than the coverage so far suggests. A huge share of everyday transactions here, UPI payments, net banking, government portal logins, happen inside a browser tab, often sitting right next to whatever else you're working on. An agent with permission to act across tabs is, by design, sitting close to that kind of data.

India's Digital Personal Data Protection (DPDP) Act puts the obligation on anyone processing personal data to collect only what's needed and to be transparent about how it's used, as we explained when the law itself was broken down in plain English. A browser extension with standing access to every open tab, installed by a user who may never read the permission prompt closely, is exactly the kind of design regulators have in mind when they talk about data minimisation. There's no indication OpenAI is out of step with Indian law here, but it's a live example of why agentic AI products are going to keep testing where that line sits.

Convenience and access tend to grow together in AI products, and users rarely get asked to choose how much of the second they want in exchange for the first.

Browser-based agents are clearly not going away; every major lab is building toward AI that acts for you rather than just talking to you. The useful habit for Indian users isn't to avoid these tools, it's to treat every "Work mode" or "agent mode" toggle as a real decision, close tabs you don't want an AI anywhere near, and actually read the permission screen before tapping "Add extension."