Per-site consent: How OpenAI limits ChatGPT's new access to your Chrome tabs
On October 15, 2026, OpenAI announced that the ChatGPT extension for Chrome can now look across open tabs, control them to complete tasks, and organize and group them. The same day, the extension was updated in the Chrome Web Store (version 1.26.1005.11759), where it is listed with 7 million users. It's a concrete expansion of OpenAI's "computer use" functionality — and it arrives in an extension many people already have installed, not in a separate browser.
What it does
The announcement splits the functionality into two tracks, as relayed by Unite.AI:
- Chat: The assistant can look across the user's open tabs to gather information and make its answers more useful. The Verge reports that tab viewing is activated by tagging Chrome in ChatGPT's web or desktop app. OpenAI published a demo video in which someone asks ChatGPT to compare the three tabs they have open with sofas they are considering buying.
- Work: ChatGPT Work, OpenAI's agent for longer and more extensive tasks, can control Chrome tabs for tasks such as sending messages, filling out forms, and downloading files.
In addition, the extension's sidebar received an upgrade with full conversation history, according to The Verge, which compares the tab-control feature to Google's Gemini-powered "auto browse" function.
In a series of posts on X from its official ChatGPT account, the company said, per Unite.AI, that it was bringing "the power of computer use to anyone who uses Chrome."
The permission model: ask per new website
Most relevant for users deciding whether to turn the features on is how OpenAI has built the permission system. OpenAI's documentation for the browser extension, as reproduced by Unite.AI, describes the following behavior:
- Ask by default. ChatGPT asks before interacting with each new website, based on the site's host.
- Four options per request: Allow once, allow for this site, allow for all sites, or decline.
- Shared lists: Users can manage permission and block domains that are shared across all supported browsers.
- Risk labeling: The documentation labels both "allow for all sites" and access to browser history as "Elevated Risk."
- History is narrowed: Access to browser history is scoped to the individual request, and there is no option to grant always-allow.
The design follows a pattern familiar from mobile platforms: granular, reversible permissions with the most invasive options labeled and partially walled off from automatic approval. "Allow for all sites" does exist as a choice, but OpenAI itself flags it as elevated risk — a signal that the company considers persistent tab access something users should think twice about before enabling permanently.
The store listing and data handling
The Chrome Web Store listing, offered by OpenAI, shows version 1.26.1005.11759, updated October 15, 2026, and reports 7 million users, according to Unite.AI. The listing also includes data-handling terms, including that data is not sold to third parties outside approved use cases.
Worth noting: the features are driven from ChatGPT's desktop app, not from the extension alone. The documentation says the extension works with Google Chrome, Microsoft Edge, Brave, Opera, and Vivaldi from the desktop app, with side chat available in Chrome, Edge, Brave, and Vivaldi — but not in Opera.
The context: after Atlas
The extension upgrades come shortly after OpenAI's own browser, ChatGPT Atlas, disappeared. Atlas was announced on October 21, 2025, and, according to a Japanese commentary post on note.com, the service was discontinued on August 9, 2026. The same post writes that OpenAI has hinted at a strategy of using the lessons from Atlas to move AI agent functions that use browsers into ChatGPT and Codex. The author stresses that OpenAI did not explain the discontinuation as having a single cause.
The Verge frames the new extension features the same way: "OpenAI's ChatGPT Atlas browser is gone, but now ChatGPT's Google Chrome extension is getting some upgrades that allow it to use the browser tabs you already have open."
That statement is worth treating with caution: the strategy of moving agent functions into ChatGPT and Codex is a secondary characterization of OpenAI's statements from a commentary source, not a documented product roadmap. But the direction is legible in the product itself: rather than asking users to switch browsers, OpenAI is moving the agent functions into the browser people already use — and thereby gets distribution through an extension with 7 million users as an installed base, instead of having to build it from zero.
What remains unresolved
As of October 15, 2026, there is no independent testing of the tab-control features, no user or expert reaction, and no security analysis of an AI agent that can read and act on websites where the user is signed in. The latter is the obvious risk a product like this raises: an agent with access to open tabs will typically encounter sessions where the user is authenticated, and the permission model handles this through per-host consent — but no one has yet examined how that works in practice.
Both outlets also cover the story via secondary reproduction of OpenAI's posts on X; neither the user count, the version number, nor the quotes from the announcement have been first-hand verified by AIMag. One source (Unite.AI) is further truncated mid-sentence about install-time permission prompts related to reading browser history, so the complete list of permissions that request consent at installation is not available from these documents.
For the 7 million users of the extension, the updated picture is nonetheless concrete: reading across tabs requires tagging Chrome in the ChatGPT app, tab control via Work requires approving each new host — or using the elevated-risk-labeled options — and history access cannot be granted permanently. How well this works outside OpenAI's demo video remains an open question.