ChatGPT Work can now control open Chrome tabs to fill forms, send messages, download files and organise tabs. The change turns OpenAI’s Chrome extension from a source of browsing context into an execution layer for tasks that already take place across websites, according to 01net’s account of the announcement.
OpenAI’s examples include completing a registration form while stopping before submission, adding a new dentist’s details and scheduling an appointment, and collecting suppliers from several tabs into a comparison document. The announcement does not specify which subscription plans or countries receive the capability first, 01net reports.
Work turns open tabs into an action surface
A user can mention the browser during a conversation in ChatGPT’s web or desktop application. ChatGPT then consults the open tabs when preparing its response. In Work mode, the assistant can go further by operating those tabs rather than merely using their contents as reference material, according to the report.
This is distinct from launching another OpenAI browser. After the company discontinued Atlas, it redistributed browser functions between its desktop application and Chrome extension. The extension could already read tabs, selected text and YouTube content; the new release adds actions within the browsing session, 01net explains.
The practical change is the distance between an answer and the next browser action. A comparison no longer has to end as prose that the user manually transfers elsewhere. Work can use the pages already open and continue into a form, message, file download or another supported browser operation.
Chat supplies context while Work executes
The two modes have different roles. Chat draws on the visible browsing context to answer a question, while Work is the mode responsible for acting on the tabs. OpenAI’s demonstration used separate product pages as inputs to a comparison, illustrating how several tabs can become a shared task context rather than isolated pages, according to 01net.
The reported controls operate around that mechanism. Access can be authorised by website, and settings include allowlists and blocklists. Consultation of browsing history requires an explicit request for permission. The redesigned side panel also places conversation history and the Chat and Work modes beside the page being viewed, the source states.
Those controls describe where the assistant may operate; they do not establish how reliably it completes every multi-step workflow. The available evidence consists of OpenAI’s stated functions and demonstrations, not independent results across different business applications. A successful action in one form therefore does not prove equivalent behaviour in another.
Conceptual illustration of the separation between browser-aware Chat and action-oriented Work.
Commercial operations gains a narrower automation route
The strongest use case is a bounded task spread across browser tabs. Supplier research is one example explicitly associated with the release: information found on several pages can be placed into a comparison document. The demonstrated expense workflow also shows ChatGPT locating a receipt in Gmail before submitting it to an expense tool, 01net reports.
Consider a hypothetical sales-operations task. A coordinator opens several event-venue pages, asks ChatGPT to compare the available information and then uses Work to enter the selected venue’s details into an enquiry form. The assistant pauses before submission, following the checkpoint shown in OpenAI’s registration example. This would reduce copying between pages without turning venue selection itself into an unattended decision.
The distinction matters because the release does not demonstrate an end-to-end commercial process running without intervention. It shows specific browser actions: reading across tabs, entering information, sending messages, arranging tabs and downloading files. Teams evaluating it should use a recurring browser task with a clear completion point, rather than treating general browser control as evidence that every web workflow has been automated.
Conceptual illustration of the browser-based receipt and expense workflow described in the announcement.
Availability and extension provenance remain open
From OpenAI’s desktop application, the extension can also control Edge, Brave, Opera and Vivaldi, according to 01net. That broadens the mechanism beyond Chrome, although the announcement cited by the publication did not clarify the initial geographic or subscription rollout.
Installation provenance is a concrete boundary for this kind of capability. Earlier in the year, two extensions impersonating ChatGPT assistants extracted conversations and browsing histories from more than 900,000 users, 01net reports. That incident does not describe OpenAI’s extension, but it makes the identity of the installed extension relevant when browser content and actions are involved.
The immediate decision is whether one repetitive, tab-based commercial task benefits from moving directly from research to browser execution. The unresolved questions are rollout availability and consistent performance across the specific sites that task requires; the stop-before-submission workflow provides the clearest initial boundary while those questions remain unanswered.
