Your pile of Chrome tabs just became AI context.
OpenAI has upgraded its Chrome integration so ChatGPT can use the pages you already have open instead of making you copy information into a prompt one tab at a time.
In normal Chat, it can look across open tabs and gather information.
In Work, it goes further.
ChatGPT can control those tabs to perform tasks such as filling forms, sending messages, downloading files and organizing the browser.
Useful?
Very.
Also a completely new level of trust to hand an AI agent?
Definitely.
Reading tabs changes the basic ChatGPT workflow
The demo OpenAI published is intentionally boring. Someone has several sofa pages open. Instead of sending each link to ChatGPT, they ask the assistant to compare what is already open.
ChatGPT can use the tabs directly. That sounds small until you think about how much modern work happens inside a browser.
Research.
Email.
CRM.
Documents.
Shopping.
Admin portals.
Dashboards.
A browser with 30 tabs is often the real working memory of the person sitting in front of it.
ChatGPT can now tap into that context.
Work can act, not just read
The bigger shift is Work.
OpenAI says Work can take actions inside Chrome tabs.
That includes things such as:
sending messages,
filling in forms,
downloading files,
navigating across tabs,
organizing and grouping tabs.
This turns Chrome from a place ChatGPT can read into an environment where an agent can operate.
That is much closer to the original vision of browser agents: stop telling the AI what is on screen and let it use the computer.
Why OpenAI is doing this inside Chrome
Instead of convincing everyone to abandon the browser they already use, ChatGPT is moving into it.
That is smart.
Chrome already contains your signed-in accounts, bookmarks, extensions, work apps, shopping sessions and all the tabs you forgot to close three days ago.
The browser is basically the operating system for a huge amount of modern knowledge work. Owning the browser is difficult. Becoming the agent inside the browser may be enough.
This also raises the stakes
Giving an AI access to public websites is one thing. Giving it access to tabs where you are already signed in is different.
Your tabs can contain Gmail, banking pages, customer data, HR tools, private documents, internal dashboards, invoices and social accounts.
OpenAI's documentation says users control which sites the browser agent can access and that ChatGPT asks before using new websites unless the user has already approved them or chosen more permissive settings.
Those controls are going to matter.
A browser agent can make a useful mistake and an expensive mistake with the same mouse button.
Prompt injection becomes more concrete
There is another problem.
Web pages are not trusted inputs. A malicious site could contain instructions designed to manipulate an AI agent that reads the page. This is known as prompt injection.
The more an agent can do after reading a webpage, the more dangerous a successful injection becomes. If ChatGPT can only summarize a page, the blast radius is limited. If it can send messages and submit forms, permissions become part of browser security.
Google now has a weird competitor inside Chrome
There is also a strategic irony here. Chrome is Google's browser. Gemini is Google's AI.
And OpenAI is now turning ChatGPT into an agent that can operate directly inside Chrome. The AI wars are increasingly happening inside products owned by competitors. The old software borders are getting messy.
Zerionia view
The important part is not that ChatGPT can “see your tabs.”
The important part is that the browser is becoming an agent interface.
Today we click the websites. Tomorrow we may increasingly tell an agent what outcome we want and watch it move across the tabs for us. That could remove huge amounts of repetitive work.
It also means a browser permission is becoming something much bigger than “can this extension read this page?”
It can now mean:
can this AI act as me here?
That is the permission worth watching.
