ChatGPT’s browser agent is the most practical AI upgrade in months. On 15 October 2026, OpenAI announced that its Chrome extension can now read the tabs you already have open — and, in a separate Work mode, take actions inside them: filling out forms, sending messages, downloading files and grouping messy tabs. No more copying and pasting between a dozen pages. This guide walks through setup, the difference between Chat and Work modes, the six best tasks to try first, and the permission checks to run before you hand it control of your browser.
What is the new ChatGPT browser agent?
The ChatGPT browser agent lives inside OpenAI’s Chrome extension and brings “computer use” into your everyday browsing. OpenAI announced it in a post from the official @ChatGPT account on X: “We are bringing the power of computer use to anyone who uses Chrome.” Instead of copying information out of websites and pasting it into a conversation, you tag Chrome in ChatGPT — by typing @ — and the assistant looks at your open tabs directly.
The context matters: OpenAI wound down its own Atlas browser project shortly before this launch, so the extension is now the main way ChatGPT reaches your browser. The Chrome Web Store lists the ChatGPT extension at roughly seven million users, which makes this the widest browser-agent rollout OpenAI has ever done — this isn’t a niche feature, it’s the default assistant arriving inside the world’s most used browser.
Chat vs Work: how the two modes differ
Chat reads your tabs; Work controls them. Everything else is detail. Here’s the split:
| Question | Chat mode | Work mode |
|---|---|---|
| What it can do | See and gather information across your open tabs | Take actions inside your tabs — fill forms, send messages, download files, organise tabs |
| How you invoke it | Mention Chrome in the ChatGPT web or desktop app | Type @ in a ChatGPT Work or Codex chat, or use the sidebar |
| Good for | Comparing prices across tabs, summarising research, pulling specs from several pages | Registration forms, contractor comparisons with draft emails, calendar adds, bulk downloads |
| Risk level | Low — read-only view of your own tabs | Higher — it acts as you on signed-in sites; always review before it submits or sends |
The side panel inside Chrome got a refresh too: it now carries your full ChatGPT conversation history, so you can ask questions in Chat or run tasks in Work right next to the page you’re reading. You can reference open tabs, ask questions about highlighted text, or ask for a summary of a YouTube video. And for anyone ending the workday with forty tabs open, you can simply ask it to organise or group them.
How to set up the ChatGPT browser agent
Setup takes about three minutes: install the extension, connect it, and confirm tab access.
- Install the ChatGPT extension from the Chrome Web Store (it’s the official OpenAI extension, already installed by around seven million users).
- Open ChatGPT on the web or in the desktop app and type @ to tag Chrome in a ChatGPT Work or Codex chat — this is how the assistant knows to look at your tabs.
- Confirm tab access. Before selecting Chrome for tab access, make sure the browser displays a Manage option — this is OpenAI’s own permission check, and it’s the control that decides which tabs the agent can see. Confirm it before proceeding.
- Alternative route: the desktop app’s settings include a Computer Use section where the browser connection can also be configured.
If you use a different browser, the extension also supports Edge, Brave, Opera and Vivaldi through the desktop app — though on Opera, the Side Chat feature isn’t supported, only tab mentions and browser control. For the full experience, Chrome is the reference setup.
Six tasks to try first (with what to expect)
Start in Chat mode — read-only first, actions later. These are OpenAI’s own demo scenarios, ordered from zero-risk to worth-a-second-look:
- Compare products across tabs. Open three product listings in separate tabs, then ask: “Compare the prices, delivery times and return policies of these three sofas.” Chat pulls the specifications from each page into one digest — the exact task that used to mean furious tab-switching and a notepad.
- Build a comparison sheet. Ask Chat to collect vendor pricing and features from your open tabs into a spreadsheet with links. Work mode can create documents, spreadsheets and slides directly from tab content.
- Summarise a YouTube video. In Side Chat, ask for a summary of the video you’re watching — it uses the captions/transcript, and you can follow up with timestamped questions.
- Ask about highlighted text. Highlight a paragraph on any page and ask ChatGPT to explain it — handy for dense documentation or legalese.
- Draft a registration email. With Work mode, ask it to compare contractor services across tabs and draft an email to the best fit — but read the draft before it goes anywhere.
- Try a form fill. Ask Work to fill out an event registration form from an event brief — and stop before submitting. OpenAI’s own demo stops at this point too: review every field, then submit it yourself.
This list is assembled from OpenAI’s own posted demo scenarios and independent reports of the launch — desk-researched, not hands-on tested, because the rollout is days old and feature details are still being confirmed across regions.
Which browsers and plans support it?
The full setup runs in Chrome, with partial support on other browsers. Independent coverage of the launch (AI Parabellum, 16 October) reports the tab-reading and control features are available on all ChatGPT plans — but OpenAI’s own guidance says to check the help centre for your region, since availability can lag outside the US. The Chrome extension’s roughly seven million users get it first; Edge, Brave, Opera and Vivaldi connect through the desktop app, with Opera missing Side Chat for now. Beyond research and form-filling, OpenAI says the agent can create or edit documents, spreadsheets and slides, enter data into business systems and download files in bulk — a sign that this is aimed squarely at repetitive admin work, not just shopping comparisons.
The risks: permission checks before you hand over control
A browser agent is powerful precisely because it acts as you — and that’s the risk. Running it well means five checks before the first real task:
- Run the Manage-option check first. Confirm the browser shows the Manage option before selecting Chrome for tab access. If it doesn’t, don’t proceed.
- Never let Work submit or send on the first run. Forms, messages and downloads should stop at review — OpenAI’s own demo stops the registration form before submission. Make that your permanent default.
- Keep it away from money and medical accounts for now. An agent acting on signed-in pages can’t fully judge page content — a prompt-injected page could steer it. The broader industry is already moving this way: major banks, Yelp and Delta all moved to block unauthorised agents from their sites in the same week this launched.
- Watch Side Chat on sensitive pages. The persistent panel is convenient, but convenience means the agent sits beside your inbox, banking and health portals. Consider which tabs are open before you invoke it.
- Verify its first answers against the tabs themselves. Spot-check one comparison against the source pages. Agents misread fine print — prices, dates and “delivery not included” footnotes — more often than demos suggest.
The honest limits: this is a first-generation rollout, and the details that matter (regional availability, exact plan gating, how permission prompts behave on sensitive sites) are still emerging. Treat anything a third party reports about the extension as provisional until OpenAI’s own help documentation confirms it.
FAQ
Is ChatGPT’s browser agent available on all ChatGPT plans?
Independent coverage of the launch reports availability on all ChatGPT plans, but OpenAI itself points users to its help centre for regional availability — confirm before relying on it.
Does the ChatGPT browser agent work outside Chrome?
Partially. Edge, Brave, Opera and Vivaldi are supported through the desktop app, but Opera misses the Side Chat feature. Chrome is the full reference experience.
Can Work mode submit forms or send messages by itself?
It can fill forms and draft messages, but you should always review before anything is submitted or sent. OpenAI’s own demo stops the registration form before submission — follow the same rule.
Is my browsing data safe with the ChatGPT browser agent?
The agent can only see tabs you grant it access to via the permission flow (look for the Manage option check). As with any agent on signed-in sites, keep it away from banking and payment pages, and spot-check its answers against the source tabs.
What happened to ChatGPT Atlas?
OpenAI’s standalone Atlas browser was wound down shortly before this launch, making the Chrome extension — now a full browser agent — the main way ChatGPT reaches your browser.
Sources and methodology
This guide was assembled on 20 October 2026 from OpenAI’s official announcement (the @ChatGPT post on X, 15 October 2026, as summarised by independent outlets) and launch-day coverage: AI Parabellum (16 October 2026), Analytics Insight (16 October 2026), Tech Featured (16 October 2026), and ExplainX (16 October 2026). Claims are desk-researched and labelled where unconfirmed; the rollout is days old and region-specific details were not hands-on tested. Related reading on this site: why banks and big platforms are blocking AI agents and our AI tool testing methodology.
Arva Rangwala covers AI news, AI tools, guides and prompts for OpenAIMaster — what is new, what is worth using, and how to put AI to work.