OpenAI has given its ChatGPT extension for Google Chrome a significant upgrade. The assistant can now read what’s open in your browser and act on it. The updated add-on, available now from the Chrome Web Store, lets ChatGPT use your open tabs as context for questions and, with permission, operate those tabs directly. That means sending messages, filling in forms, downloading files, and grouping a cluttered tab bar into something manageable.
The new sidebar is the more visible change, and it brings ChatGPT closer to how Gemini already works inside Chrome. Instead of copying text into a separate chat window, you ask questions alongside the page you’re looking at. OpenAI’s examples are deliberately ordinary: comparing three couches across dimensions, materials, delivery times and return policies, summarising a long article’s central argument, or checking a stack of software pricing pages against a list of requirements.
The agentic features matter more. OpenAI describes ChatGPT completing a registration form from an event brief and stopping before it submits, moving vendor details from several tabs into a comparison spreadsheet, and filling out a new-patient form for a dentist before adding the appointment to two separate calendars. These are the repetitive, multi-tab tasks that eat up time online, and they are what browser agents have been promising for over a year.
Strategically, this is OpenAI building inside a competitor’s browser. Google controls Chrome and is steadily embedding Gemini into it, while OpenAI already has its own browser, Atlas, aimed at the same AI-driven browsing use cases. Shipping a capable extension acknowledges that most people aren’t going to switch browsers to try an assistant. It also leaves OpenAI dependent on Google’s extension rules and on whatever room Google allows for rival AI tools on its platform.
There are reasons for caution. An assistant that can read every open tab and submit forms has access to email, banking dashboards and work documents, and AI agents acting on web pages remain exposed to manipulation through hidden instructions on those pages. OpenAI’s own example of stopping before a final submission suggests the company knows confirmation steps matter. How consistently that applies across sensitive actions isn’t clear from the announcement, and neither is whether the full feature set is limited to paid ChatGPT tiers.
The update follows OpenAI’s recent work on always-on agents and its move into advertising inside ChatGPT. The aim is plainly to make ChatGPT the layer people work through, not just a site they visit. Whether users trust it with their logged-in tabs is the open question.
