ChatGPT can take care of web tasks without your password
OpenAI's ChatGPT Work can now use your websites for you, with a cloud browser that keeps your passwords hidden from the AI.
OpenAI has expanded ChatGPT Work with a cloud browser that can open websites, navigate pages, fill in forms and complete certain online tasks on a user’s behalf.
The most impressive part is that it can also work with signed-in websites without exposing the user’s password to the AI model! Here's everything you need to know about this latest feature!
ChatGPT gets its own browser
The cloud browser runs on a separate remote computer rather than using the browser on the user’s device.
That means it does not access open tabs, saved passwords, cookies, extensions or browsing history from the personal device.
Instead, ChatGPT works inside its own browser session, where it can read webpages, click buttons and enter details when supported. That could help with tasks such as checking restaurant availability, comparing product stock, tracking packages, preparing bookings, finding appointments or working with business tools.
Exactly what ChatGPT can complete will still depend on the website, the user’s access and the task.
Your password stays out of the chat
When ChatGPT reaches a supported login page, it pauses and asks the user to sign in through a secure form.
The user enters their username, password, and any two-factor authentication code themselves.
OpenAI says these credentials go directly to the remote browser, meaning they are not visible to the model and are not stored by ChatGPT. Once signed in, GPT can continue working within that session. The session may remain active until it expires or the user clears the browser data.
So users may not have to log in repeatedly, but they still need to keep an eye on what the AI is doing.
The useful bit is what happens after login
A login screen has traditionally been a major roadblock for AI agents. With signed-in website support, ChatGPT Work can move further into real workflows, whether that involves online dashboards, travel portals, utility accounts or workplace systems.
OpenAI also says ChatGPT is designed to ask for confirmation before important actions that could create financial, legal, account-related or other real-world commitments. In other words, the AI can help get things ready, but users should still have the final say when the stakes are higher.
There are still a few catches
Giving an AI access to websites also creates new risks. It could encounter sensitive documents, emails, account settings or misleading webpage content. Users should therefore avoid entering passwords, security codes or payment details directly into the chat.
Sensitive information should be entered through the secure sign-in flow or takeover mode. Some websites may also block automated browser agents, while certain tasks may still require the user to take over and complete the final step. Users can manage website permissions and clear cloud browser data when needed.
ChatGPT is getting better at doing, not just answering
AI assistants are starting to move beyond telling users what to do and actually doing parts of the job. ChatGPT can now navigate more of the web on a user’s behalf, including behind supported login screens.
But for anything sensitive, the safest approach is still simple: let the AI do the legwork, but keep your hand on the steering wheel.