Ctrl+C, switch windows, hit Ctrl+V, and pray the formatting survived the trip. OpenAI wants to retire that workflow. On October 15, 2026, the company rolled out a major upgrade to its official ChatGPT Chrome extension, bringing cross-tab awareness, a persistent conversation sidebar, and direct browser task automation to the browser where most PC users already live.
Instead of asking users to abandon their daily browser for a standalone AI wrapper—a path OpenAI stepped back from after discontinuing its standalone ChatGPT Atlas browser earlier in 2026—this update brings "computer use" capabilities straight into Chromium. With version 1.26.1005.11759 now live in the Chrome Web Store for an install base of roughly 7 million users, ChatGPT can inspect open tabs, compare live web data, and actively click, type, and navigate across websites on your behalf.
For Windows power users, developers, and enterprise IT administrators, the update is equal parts productivity booster and security headache. Giving an AI assistant eyes on your authenticated browser sessions changes how day-to-day web tasks work—and raises immediate questions about where your session data goes.
Chat vs. Work: What the Upgraded Extension Actually Does
OpenAI splits the new browser capabilities into two distinct operational modes: reading context in Chat and taking action through ChatGPT Work (or Codex chats in the desktop workflow).
Cross-Tab Context with @chrome
In standard Chat mode, you no longer need to feed URLs or pasted text to the model one tab at a time. By mentioning @chrome in ChatGPT, you grant the assistant permission to look across your open browser tabs and synthesize what it finds.
In OpenAI's demonstration of the feature, a user opened three separate retail tabs for sofas and asked ChatGPT to compare them. The assistant pulled product specifications, pricing, and dimensions across all three tabs without the user copying a single line of text. Beyond shopping, the practical applications for PC users are obvious: comparing documentation pages, cross-referencing cloud pricing tiers, or pulling numbers from multiple internal dashboards at once.
Browser Automation with ChatGPT Work
While Chat mode reads, ChatGPT Work acts. Integrated into the extension and desktop workflow, Work allows the AI to interact directly with web page controls across multiple tabs. According to OpenAI's documentation and demonstrations, supported automation tasks include:
- Vendor and Contractor Research: Collecting quotes, comparing feature matrices across tabs, and organizing the findings into a structured spreadsheet.
- Form Completion: Pulling personal or business details from local documents to fill out event registrations or new-patient forms, with the option to pause so the user can review the fields before hitting submit.
- Communications and Scheduling: Drafting emails to selected service providers, sending messages, downloading files in bulk, and placing appointments onto personal or work calendars.
- Tab Housekeeping: Grouping, organizing, and cleaning up sprawling rows of open browser tabs.
- Authenticated Dashboard Actions: According to OpenAI's training documentation, the tool can operate inside sites where you are already signed in—such as Salesforce, LinkedIn, Gmail, or internal company portals—adjusting dashboard filters, opening reports, and entering records.
The Persistent Side Chat Panel
The update also overhauls the in-browser sidebar. Clicking the ChatGPT icon in the browser toolbar opens a side panel anchored to the current tab. Unlike basic web clippers, this panel carries your full ChatGPT conversation history and supports both Chat and Work modes side-by-side with the active webpage. If you start a conversation in the browser sidebar, it syncs immediately with the main ChatGPT app, allowing you to move between the browser panel and the desktop interface without losing your place.
Section Summary: The updated extension moves ChatGPT from a passive Q&A box to an active browser operator, combining @chrome multi-tab reading, a synced history sidebar, and multi-step form and workflow automation via ChatGPT Work.
Not Just Chrome: Chromium Support and Setup Boundaries
Although the headline announcement centers on Google Chrome, Windows users running other Chromium-based browsers are not left out in the cold. OpenAI's official browser extension documentation confirms support across five major browsers, though feature parity varies slightly:
| Browser | @ Tab Mentions & Browser Control | Side Chat Panel Support | Notes |
|---|---|---|---|
| Google Chrome | Yes | Yes | Primary target; full support |
| Microsoft Edge | Yes | Yes | Compatible via Chrome Web Store |
| Brave | Yes | Yes | Full desktop and sidebar support |
| Vivaldi | Yes | Yes | Supported in current OpenAI documentation |
| Opera | Yes | No | Tasks must be launched from the ChatGPT desktop app |
There are also important platform and account boundaries to keep in mind. While Windows Report and The Verge note that @chrome can be tagged from the web or desktop app, OpenAI's setup documentation emphasizes the ChatGPT desktop app (Settings > Computer Use) as the control hub for configuring full browser control, managing site allowlists, and running Work or Codex automation sessions. Furthermore, as reported by PiunikaWeb, access to ChatGPT's Work mode requires at minimum a paid ChatGPT Plus subscription, meaning free-tier users will not get full autonomous browser control.
It is also worth distinguishing @chrome from @Browser. In the ChatGPT desktop app, selecting @Browser launches an isolated, built-in browser inside ChatGPT itself. Selecting @chrome (or your configured Chromium browser) bridges ChatGPT directly into your regular browser profile, complete with your active cookies, extensions, and logged-in sessions.
How to Set Up Browser Control
According to OpenAI's documentation, setting up the full desktop-to-browser workflow follows a specific sequence:
- Update the ChatGPT desktop app to the latest version and ensure you are using the browser profile where you want the extension to live.
- Open the desktop app and navigate to Settings > Computer Use.
- Select your target browser (such as Chrome or Edge) and install the ChatGPT browser extension from the Chrome Web Store (
Version 1.26.1005.11759, 1.07 MiB) when prompted. - Approve the browser permission prompts, then return to the ChatGPT desktop settings to verify that the button next to your browser now reads Manage instead of Install.
- Start a new Work or Codex chat and type
@to select your browser from the mention menu. - Optional: If your automation workflow requires uploading local files from your PC into a Chrome tab, open
chrome://extensions, click Details on the ChatGPT extension, and toggle on Allow access to file URLs.
If the connection fails to initialize, OpenAI recommends verifying that the target website is not on your blocklist, confirming the extension is enabled in the active browser profile, and restarting both the browser and the ChatGPT desktop app.
Section Summary: Despite the Chrome branding, the extension works in Microsoft Edge, Brave, Vivaldi, and (without side chat) Opera. Full setup and permission management run through the Computer Use menu in the ChatGPT desktop app.
The Security and Privacy Reality Check
Whenever an AI agent is handed the keys to an authenticated browser session, IT administrators and security-conscious users should pause and read the fine print. Because the extension operates inside your normal browser profile, it inherits the access rights of whatever accounts you have open—from corporate webmail to cloud admin consoles.
OpenAI's documentation outlines the technical permissions the extension requests in Chrome, and the list is extensive:
- Access to the page debugger
- Reading and changing data on visited websites
- Accessing browser history, bookmarks, downloads, and notifications
- Managing tab groups
- Communicating with cooperating native applications on the host computer
In the Chrome Web Store's privacy disclosures, OpenAI declares that the extension may handle personally identifiable information, health data, financial and payment info, authentication information, personal communications, location data, web history, user activity, and website content—while stating that this data is not sold to third parties or used for creditworthiness decisions.
Guardrails, Prompt Injection, and History Access
To keep the assistant from going rogue across your open tabs, OpenAI built several permission gates into the workflow:
- Per-Host Confirmation: By default, ChatGPT asks for permission before interacting with each new website, identifying the target by host (for example,
example.com). You can approve access once, allow that specific host for future tasks, decline, or enable "allow all sites." OpenAI explicitly warns that allowing all sites removes confirmation prompts and carries elevated risk. - Allowlists and Blocklists: Under
Settings > Computer Use > Managein the desktop app, users can pre-configure domain allowlists and blocklists that apply across all supported browsers. - Scoped Browser History: Accessing your browsing history requires a separate permission prompt. OpenAI notes that history often contains sensitive internal URLs and search queries, so history access is strictly scoped to the single request—there is no "always allow" toggle for browser history.
- Data Retention: OpenAI states it does not record a separate, continuous log of every browser action from the extension. However, once page text, screenshots, summaries, or tool calls are pulled into a task, they become part of your ChatGPT conversation context and are governed by your account's ChatGPT data controls.
The biggest architectural risk in browser automation is indirect prompt injection. If ChatGPT reads a webpage containing hidden or malicious text designed to hijack AI instructions, an unmonitored agent could be tricked into copying sensitive context or submitting unintended forms. OpenAI's own documentation warns users to treat all webpage content, text selections, and transcripts as untrusted context, advising users not to pass secrets or highly sensitive records through browser tasks unless necessary—and to stay at the keyboard to review prompts before allowing ChatGPT to submit forms or execute actions.
For Windows Forum readers who manage company networks or simply value a locked-down PC, the takeaway is straightforward: treat an AI browser extension like an intern sitting at your unlocked workstation. Use separate browser profiles to isolate personal or sensitive banking tabs from AI-assisted work tabs, never enable the blanket "allow all sites" toggle, and let ChatGPT draft the form—not click the final submit button.