ChatGPT Just Moved Into Chrome to Take on Google's Browser

By Daniel Rao

San Francisco, October 17, 2026

OpenAI spent the better part of a year building its own browser. Then it killed it. On Thursday, the company finally admitted what the traffic numbers had been screaming all along: nobody is going to switch browsers for ChatGPT — so ChatGPT is moving into the browser everyone already uses.

The announcement arrived as a thread from the official ChatGPT account on X, timestamped October 15. "A few big upgrades for Chrome users today! We're bringing the power of computer use to anyone who uses Chrome," the first post reads. Within hours it had pulled 1.2 million views. The substance is bigger than the fanfare suggests: ChatGPT can now look across every tab you have open in Chrome, and — in its Work mode — reach into those tabs and act. Send messages. Fill out forms. Download files. Tidy your tab chaos into groups.

Read that again. An AI assistant that can click and type inside the browser where you are signed in to everything.

Photo: Google Chrome running with several tabs open (file photo) — the territory ChatGPT just moved into. Credit: Lordalpha1 / Wikimedia Commons (CC BY-SA 4.0).

What shipped: Chat reads, Work acts

OpenAI drew a deliberate line between its two modes. "If you're in Chat, it can look across your open tabs and gather info," the announcement said. Chat is retrieval: the assistant pulls text from pages already loaded and uses it to answer. Ask it to compare the three sofas you have open by dimensions, material, delivery time and return policy, and it does the reading for you.

The consequential half sits in the second mode. "If you're in Work, ChatGPT can control your Chrome tabs to get things, like sending messages, filling out forms, or downloading files. It can even help you organize and group that pile of tabs you have open," the thread continued. That is action, not retrieval: the assistant clicks, types and submits inside pages where you are typically already signed in.

The demo prompts OpenAI published make the ambition plain. "Use the details in this event brief to fill out the registration form, and stop before submitting." "Put the vendors in these tabs into my comparison sheet, with pricing, key features, and source links." "Fill out the form I have open in @chrome ahead of seeing my new dentist. Also, can you add the appointment to my work and personal calendar." This is not a chatbot that answers questions about the web. It is a stand-in that does web errands.

There is also a redesigned surface: the full power of ChatGPT now sits in a side panel inside Chrome, your entire conversation history alongside the page you are viewing. The Chrome Web Store listing describes the whole product in five words: "Control your browser with ChatGPT." You do not get more honest than that.

Ben Schoon, senior editor at 9to5Google, covered the update on October 16 and put it crisply: OpenAI "announced deeper integration between ChatGPT and Google Chrome, including the ability to hand control over your tabs to the AI assistant." His piece notes the new sidebar works much like Chrome's own built-in Gemini integration — which is precisely the comparison that should worry Google.

Atlas died so this could live

To understand why this matters, rewind eleven months. On October 21, 2025, OpenAI launched Atlas, a Chromium-based browser of its own with an agent mode for Plus and Pro subscribers. The pitch was the classic one: switch browsers, get the AI-native future. Users declined. By May 2026, HUMAN Security's measurements put Perplexity's Comet at roughly 47% of agentic web traffic — and Atlas at about 20.3%. Less than nine months after launch, on July 9, 2026, OpenAI discontinued Atlas outright and folded its browsing functions into ChatGPT Work, an agent built on GPT-5.6.

The Chrome extension is the admission and the fix in one move. Chrome holds about 66.3% of global web traffic, per Cloudflare's Q3 2025 numbers. Persuading two-thirds of the web to move house was never going to work; moving into the house was always the smarter play. OpenAI tried the hard thing, failed at it, and has now done the obvious thing. There is something almost refreshing about the bluntness of the pivot — and something a little alarming about how fast the company cycled through a flagship browser to get here.

This is also a direct assault on Google's home turf, and Google knows the shape of this fight. Google began building Gemini into Chrome in September 2025, including analysis across multiple open tabs, and in July 2026 added a persistent Gemini 3 side panel plus an auto-browse agent for its US AI Pro and Ultra subscribers — one Google said would pause for confirmation before purchases or social posts. Now both assistants offer a side panel that sees your tabs and acts in them. The difference: Gemini ships with the browser. ChatGPT arrives as an extension you install from Google's own Web Store. Google is now hosting its most aggressive competitor inside its own product.

The permissions are the story

Before the excitement, read the install prompt. Chrome's listing for the extension requests broad access: data on all websites, browsing history, downloads and tab groups. That is the price of admission for an agent that works across your tabs, and it is worth saying plainly what it means — the assistant can see the tabs you grant it access to, which includes, in principle, the tab where your bank is open. Asked whether ChatGPT can view banking tabs, the honest answer from the coverage is: it can read the tabs you grant it access to, and it asks permission for each new site. Shut the sensitive tabs before you use it.

OpenAI says the agent asks before interacting with a new website, and that by default it requests permission per site. That is the right design. But here is where the story gets mushy, and where we did something the launch-day coverage mostly skipped: we read OpenAI's thread and 9to5Google's report back to back, and they do not quite agree on who gets what. OpenAI's thread closes with "Available for all ChatGPT plans" — but the line sits directly under the extension download link, reading as a statement about the extension rather than every feature inside it. When OpenAI folded Atlas into ChatGPT Work in July, Work began rolling out to Pro, Enterprise and Edu subscribers first, with Plus and Business following. Neither the October thread nor the secondary reporting says whether free users get Work's tab-control inside Chrome, or only the Chat side that reads tabs. If you are on the free tier, check your own account before assuming the demo applies to you.

The security hole nobody has closed

Now the uncomfortable part. A browser agent that types into signed-in tabs inherits your access, your sessions, your logged-in everything. That layer is exactly where prompt injection lives: hidden text on a webpage that the agent reads as an instruction. When Anthropic opened its own Claude for Chrome to a research preview of 1,000 Max subscribers in August 2025, it disclosed a 23.6% attack success rate across 123 test cases without mitigations, falling to 11.2% with them. Brave and LayerX spent August to October 2025 disclosing prompt-injection flaws in Perplexity's Comet browser, and Perplexity's own security team acknowledged that malicious prompt injection remained unsolved across the industry.

OpenAI's October thread publishes no comparable figures. It says nothing about confirmation prompts before the assistant sends a message or submits a form. One of OpenAI's own sample prompts tells the agent to fill a registration form "and stop before submitting" — which suggests the stopping is something the user specifies in the request, not something the product guarantees on its own. An agent operating inside already-authenticated tabs works at precisely the layer attackers target with session hijacking, and the scope of what Work can touch — and how it is constrained — is the question security teams will ask before marketing teams do. OpenAI has shipped the capability. The safety accounting is, for now, a blank page.

Who loses when the assistant does the browsing

There is a second-order story here that website owners should not miss. Several of OpenAI's sample prompts land squarely on commercial pages: compare pricing pages against requirements, rank B2B vendors into a spreadsheet, fill a dentist's intake form. A form submitted by ChatGPT in Work arrives from your own Chrome profile, with your cookies and your fingerprint — requested by a real customer, executed by software. Whether that traffic is distinguishable from a human session, and whether it should be, is unaddressed. Publishers face their own version: "Explain this article in plain English. What's the main argument, and what evidence supports it?" is one of OpenAI's own examples. The page still loads, the ads may still count, but the reading moves to a summary in the side panel. How much attention the article gets when an assistant is summarizing alongside it is something neither OpenAI nor its coverers have touched.

And there is the commercial thread underneath. ChatGPT's ad business crossed a $1 billion annualized run rate by late August 2026, with about 1,200 US advertisers spending in August alone. Nobody says whether ads appear in the Chrome side panel. But more time spent inside ChatGPT while browsing other people's websites is more time inside an environment where OpenAI now sells ads — a neat trick, if you think about it: the assistant that reads the web for you, monetized by the company that built the assistant.

So where does this leave us? My read: this is the smartest browser move OpenAI has made, precisely because it stops pretending users will change behavior. Atlas was the industry's standard delusion — that a better product wins distribution. Distribution won, as it always does, and OpenAI has now rented it from Google instead of building it. That is colossally pragmatic. But the company shipped an agent that can act inside your signed-in life while publishing zero safety figures, leaving plan eligibility fuzzy and the confirmation model vague. Pragmatism about distribution does not excuse vagueness about control. Whether ChatGPT-in-Chrome becomes the default way people use the web will be decided less by the demos — which are genuinely impressive — than by the first ugly incident in which the agent does exactly what a webpage told it to, and nobody can say who is responsible.

FAQ: ChatGPT in Chrome

Do I need a paid ChatGPT plan to control tabs?
OpenAI says the extension is "available for all ChatGPT plans," but the tab-control half lives in Work mode, which launched for paid tiers in July. OpenAI has not clarified whether free users get Work inside Chrome or only the tab-reading Chat side. Check your account.

Can ChatGPT see my banking tabs?
It can read the tabs you grant it access to, and it asks permission for each new site. If you would not want an assistant reading a tab, close that tab before invoking it.

Does it work in browsers other than Chrome?
Yes — Edge, Brave, Opera and Vivaldi are supported through the desktop app plus extension, though the side chat panel is not offered in Opera.

What happened to Atlas, OpenAI's own browser?
Launched October 2025, discontinued July 2026 after trailing Perplexity's Comet badly in agentic traffic. Its browsing functions were folded into ChatGPT Work. The Chrome extension is effectively Atlas's second life — without asking you to switch browsers.

Related on TechPulse USA: Google's AI answers finally link back to publishers — the other front of the Google-OpenAI war; OpenAI's EU answer-labeling regime; and what safety tests found in ChatGPT for teens.

Sources: OpenAI (official ChatGPT account thread on X, October 15, 2026); 9to5Google (Ben Schoon, October 16); PPC Land; EM360Tech; Unite.AI; Gadgets Now (Indiatimes).

Daniel Rao is a technology correspondent for TechPulse USA, covering AI products, platform power, and the economics of the web. He has written about browser wars since the first one.