OpenAI has updated the ChatGPT extension for Chrome, expanding its ability to bring the tabs you already have open into conversations and tasks. The Chrome Web Store lists the update as dated October 15. Beyond gathering information from multiple tabs, ChatGPT Work can now carry out tasks such as data entry in the browser. A conversation panel beside the page also lets you use ChatGPT without leaving what you are reading, so browsing and requests can continue on the same screen. Because the feature handles pages where you are already logged in, however, it helps to think separately about what you let ChatGPT read and which actions you let it perform.
From comparing open tabs to getting work done
Through the extension, ChatGPT can use information from your open tabs in its answers. For example, you could open candidate products in separate tabs and ask it to compare prices and specifications side by side. The connection to the browser is meant to cut the effort of moving between pages and copying content into the chat box yourself.
ChatGPT Work, which handles task requests, goes on to act in the browser. The uses OpenAI lists in the store include entering data into systems of record and creating or editing documents, and it says the extension also supports bulk file downloads and tab organization. This moves ChatGPT from reading an answer and stopping to getting work done on the screens you have open.
The extension can also be used on sites where you are already logged in. The official guide gives Gmail, Salesforce, and internal tools as examples. Since this covers work that relies on existing login sessions, not just research on public pages, you should be mindful of what information and which changes that account can access if you hand over an internal input screen.
Side chat is a conversation panel that opens ChatGPT next to the page. A conversation started there can be continued in the ChatGPT app, and recent conversations can be opened from the panel. In addition to whole tabs, you can ask questions about text you select on a page.
The store description also says the extension can use installed plugins and conversation history, and, if connected, local files. OpenAI says processing in the browser runs in the background without taking over your browsing. These descriptions, however, do not tell us how often tasks succeed on individual sites or how long they take.
Reading permissions and storage in four parts
Chrome extension permissions, per-site approval, confirmation for history and file actions, and storage of data that enters a conversation are all handled separately. Using OpenAI's official guide and the store description, both checked on October 16, 2026, and organizing them by what you are granting and the stage of processing, we get the following.
| Stage | Targets and behavior in the official description | Point to keep distinct |
|---|---|---|
| Installing the extension | Chrome may ask for permissions such as reading and changing data on all sites and access to the debugger | Permissions needed to run the browser. ChatGPT also has its own confirmations at run time |
| Per-site use | By default, asks before using a new site. Options: allow once, always allow the site, allow all sites, or deny | For a site you always allow, confirmation can be skipped on later use |
| History and file actions | Asks before using browsing history, and history has no always-allow option. Downloads and uploads also reportedly require permission | Don't treat a site's standing approval as the same as confirmation for history or files |
| Storage of data entering a conversation | Text read from pages, screenshots, and similar content become subject to storage once they enter ChatGPT's context | The statement that no separate, complete record of actions is saved does not mean information used in a conversation isn't retained |
This table reflects the categories in the official descriptions and is not the result of auditing the implementation. It matches the guide's explanations of permissions and site use with its explanations of history and stored data. The confirmation for file actions is supplemented by the store's wording. Separating who you grant permission to from how information is handled afterward makes it harder to misread what you have actually approved.
For instance, enabling ongoing use of a work site does not mean you have granted standing permission to consult browsing history. On the other hand, if you choose to allow all sites, confirmation before using a site goes away. In the official descriptions we checked, there is no guarantee that every action, such as submitting a form on an approved site, will prompt for reconfirmation.
The explanation about storage also bears directly on usage decisions. OpenAI says the extension does not save a separate, complete record of browser actions, but that activity that enters the conversation context is stored. This includes not only page text but also screenshots and tool calls, so it can't be read as meaning that nothing is retained because you are just using the browser.
Content processed in a conversation is subject to ChatGPT's data control settings. In other words, you need to decide both which sites to allow and whether the information read there may be included in a conversation. The convenience of working in logged-in screens also means the tool may handle highly sensitive information.
Requirements to get started, and how much to delegate
To operate Chrome from the desktop version, update the ChatGPT app, select the browser under "Computer Use" in settings, and connect the extension. The official guide says to confirm that "Manage" appears after connecting, then mention the browser with "@" in a Work or Codex conversation. You must also use it in the browser profile where the extension is installed.
It supports browsers other than Chrome, with differences in what is available. According to the official guide, side chat works in Edge, Brave, and Vivaldi, while Opera supports tab reading and control from the desktop but not side chat. Availability can also depend on the pace of rollout and workspace settings, so the extension's presence in the store alone does not mean every feature is available in your environment.
For work requests, separating "turn the open candidates into a comparison table and keep source links" from "fill out the application form for the chosen candidate" makes it easier to check results along the way. If you hand over data entry, one approach is to ask it to stop before submitting and review the contents yourself. That is a way for users to specify the scope of work, and should be distinguished from any guarantee that the feature will always stop before every submission.
OpenAI also urges treating page content as untrusted input. Even on a legitimate site you chose, text on the page is not necessarily an appropriate instruction for the AI, and the guide mentions the risk that malicious content could cause information such as history to be copied somewhere unintended. Allowing access to a site is a separate matter from whether instructions written there should be followed.
As Chrome integration widens the work you can delegate, it matters more to specify exactly what ChatGPT should read and what it may change. If you build a workflow that checks sources in comparison tables and verifies entered data before submission, the extension can serve as an entry point for turning familiar tabs from research into practical work.