ChatGPT can now fill out forms and send messages in your Chrome tabs
OpenAI has upgraded its ChatGPT extension for Chrome so the assistant can read open tabs, send messages, fill out forms and download files — capabilities that previously lived in the now-retired Atlas browser. The update was announced on Thursday in the week of October 15–16, 2026, roughly two months after OpenAI's own browser stopped working, and in the middle of Google's competitive push with Gemini features in Chrome.
What's new
The announcement came in a thread on X from OpenAI's official ChatGPT account. "We're bringing the power of computer use to anyone who uses Chrome," the company wrote, as quoted by IBTimes (339056ed).
The update gives the extension three main capabilities, according to OpenAI as reported by The Decoder (74ab48b2):
- @chrome tagging: In ChatGPT on web or desktop, the user can tag Chrome to give the assistant visibility into open tabs. The Verge relayed a demo video in which someone asks ChatGPT to compare three open tabs of sofas they are considering buying (e53b10c8).
- Chat mode: Pulls data from multiple tabs at once — for example, comparing products across stores by size, material, shipping time and return policies, summarizing articles in plain language, or matching price pages from different sellers, according to OpenAI's own examples as relayed by The Decoder.
- Work mode: Here ChatGPT controls the tabs themselves through Computer Use. That means the assistant can send messages, fill out forms, download files, and organize and group tabs, according to IBTimes (339056ed) and The Verge.
In addition, the side panel has been upgraded: the full ChatGPT conversation history now sits alongside the web page you're on (74ab48b2).
9to5Google, which reported from OpenAI's X thread, highlights one detailed example: when ChatGPT fills out a registration form, it is instructed to stop before submitting — that is, to leave the final decision to the user (e4174a1c).
How to set it up
The control features require a bit more than just the extension. According to OpenAI's documentation, as relayed by IBTimes UK (496c7da6), the user must:
- Select their browser under Computer Use in the ChatGPT desktop app
- Install the extension
- Review the relevant permissions
By default, ChatGPT is supposed to ask before interacting with a new website. The extension supports Chrome, Edge, Brave, Opera and Vivaldi. Side chat is available in Chrome, Edge, Brave and Vivaldi — Opera users have to start browser tasks via the desktop app instead.
One important detail: the extension can work on websites where the user is already signed in, including email services, CRM platforms and other workplace tools (496c7da6). This is what makes the features useful for actual work — and what makes the security questions real.
OpenAI says the extension is available on all ChatGPT subscription plans. The Chrome Web Store listing shows around 7 million users, but that figure comes from the listing itself and has not been independently verified (339056ed). It is also unclear exactly where the plan boundaries fall for Work mode — the individual sources are not fully aligned on this, and rollout status may vary.
From its own browser to an extension
The backstory is as interesting as the update itself. OpenAI launched ChatGPT Atlas, its own browser, in October 2025. In March 2026, the company said it would merge Atlas, the ChatGPT desktop app and the coding tool Codex into a single application. In July came word that Atlas would be discontinued, and on August 9 the browser stopped working (339056ed).
Now the Computer Use capabilities from Atlas have been ported over to the world's most-used browser. Engadget describes the update as something that makes ChatGPT a more genuine alternative to Gemini in the browser (1f998ee5) — and The Verge points to the similarity with Google's Gemini-powered "auto browse" feature (e53b10c8).
The approach says something about a broader shift: rather than getting users to switch browsers, AI agents are being embedded as extensions in the browser people already use. For Google, which is itself rolling out Gemini in Chrome, that means its competitor now sits directly on its own platform.
The security caveats
Security researchers have long warned that AI agents browsing the web can be vulnerable to manipulation — for example through hidden instructions on web pages designed to trick the agent into performing unwanted actions, as relayed by IBTimes (339056ed). These are general warnings about so-called prompt injection attacks, not documented incidents tied to this extension.
But the risk is concrete given the feature set: an agent that can act on pages where the user is signed in — email, CRM, workplace tools — has access to environments where a successful manipulation could have serious consequences. The ask-before-acting default and the stop-before-submitting example in OpenAI's demos suggest the company has built in brakes, but there has been no independent testing of how robust they are.
What's verified — and what isn't
All of the capability claims in this story come from OpenAI itself, relayed through secondary reporting by The Verge, The Decoder, 9to5Google, IBTimes and Engadget. None of the outlets has independently tested the extension. The user figure of around 7 million comes from the Chrome Web Store listing. It also remains unclear how far the rollout has progressed and exactly which subscription limits apply to Work mode.
The open question over the coming weeks will be whether the features work as well in practice as in OpenAI's demos — and whether Google's response will be to tighten the conditions for third-party extensions with this kind of access to Chrome.