The ChatGPT extension for Chrome can now read and control open tabs
IA en un minuto newsroom · Editor: Jon Elgezabal
In 30 seconds
OpenAI has expanded what its ChatGPT extension does inside Chrome: it now looks at open tabs to give better answers and, in Work, it also operates them on the user's behalf. It can send messages, fill out forms or download files without leaving the browser. The extension asks for confirmation before acting on each new site and has 7,000,000 users.
OpenAI announced on Thursday, October 15, in a post on X, that the ChatGPT extension for Chrome can now use the tabs open in the browser. In Chat mode, the assistant goes through those tabs to gather information and sharpen its answers. In Work, the agent OpenAI reserves for long and complex tasks, it can also operate them: sending messages, completing forms or pulling down files. The company adds that it can also help sort and group the heap of tabs each user piles up.
According to its Chrome Web Store listing, the extension has 7,000,000 users and its latest version dates from October 15. Unite.AI reports that OpenAI offers it on every ChatGPT plan.
OpenAI's documentation explains that, by default, ChatGPT requests confirmation the first time it is about to interact with each website, which it identifies by its domain. Users can keep one list of permitted domains and another of blocked ones, or authorize it everywhere, an option the documentation itself rates as elevated risk. It gives the same label to access to browser history, which cannot be left permanently authorized. The store listing adds that the extension always requests permission before looking at history and before downloading or uploading files.
The extension also runs in Microsoft Edge, Brave, Opera and Vivaldi from the ChatGPT desktop app. It is the path OpenAI pointed its users to when it announced in July that it was retiring Atlas, its own browser, with the shutdown set for August 9, according to the ChatGPT release notes cited by Unite.AI. The documentation warns that availability may depend on the rollout and on each workspace's settings.
Why it matters · analysis and opinion
The leap is not in reading pages, which assistants already did, but in acting with each person's open sessions: their email, Salesforce or the company's internal tools. That turns every permission into a security decision. OpenAI has chosen to ask site by site and to flag the blanket permission as risky, and that design leaves the final say to the user: anyone who grants everything to move faster exposes the agent to whatever a malicious page says. After shutting down Atlas, OpenAI's bet is to get into the browser people already use instead of asking them to switch. In a company, the prudent move is to decide in advance which domains go on the allowlist and which on the blocklist, rather than leaving it to each employee.
Source: Unite.AI · Written with the help of AI: how we make the news
