ChatGPT can now things for you without seeing your logins
ChatGPT can now things on your behalf—checking insurance, canceling flights, and rescheduling deliveries—without seeing your username or password. OpenAI’s ChatGPT Work uses a secure cloud browser and form so credentials stay hidden from the model, available to Plus, Pro, and Business users on mobile and web.
That is the big shift: the assistant no longer stops at advice or in-chat bookings. It can step into the same signed-in websites you would open yourself, then finish everyday errands while keeping login details out of the model’s view. For anyone who remembers pasting passwords into chat windows, this lands as a clear then-and-now moment in how we trust tools with our accounts.
OpenAI announced the change on X. According to coverage from Mashable, ChatGPT Work—powered by GPT‑5.6—pulls those workflows into one place inside the chatbot. The pitch is practical life admin, not only boardroom tasks: utilities setup, passport or medical appointments, plane tickets, and package timing.
Key Takeaways
- ChatGPT Work can complete signed-in tasks such as insurance checks, flight changes, and delivery reschedules without the model seeing your login details.
- You enter credentials in a secure form inside a remote cloud browser; OpenAI says those details are not visible to the model and are not stored.
- The feature is available on mobile and web for Plus, Pro, and Business users.
- ChatGPT asks for confirmation before hard-to-reverse actions such as bookings or payments.
- You control site access under Settings > Cloud browser, though some sites may block AI agents.
What can ChatGPT actually do for you now?
In the past, getting an AI to “help” with a real account usually meant copying steps by hand—or worse, sharing a password in plain chat. That habit was always risky. The new flow is built to replace that brittle, unsafe pattern.
Once you start a task and describe what you need, ChatGPT can pause the conversation and present a secure login form through a remote browser. You may need to approve the site and finish two-factor authentication yourself. After that, the chat resumes and the agent continues the work.
Reported examples include checking insurance, canceling flights, rescheduling deliveries, setting up home utilities, booking passport or medical appointments, buying or canceling plane tickets, and rearranging when a package arrives. That list moves ChatGPT past light in-chat booking into deeper purchasing and account work.
It still asks before actions that could be hard to reverse or that create financial, legal, account, or other real-world commitments—such as confirming a booking or making a payment. You keep the final say when money or lasting changes are on the line.
How does ChatGPT avoid seeing your login details?
The safeguard sits in the architecture, not in a polite promise. OpenAI’s help material, as reported by Mashable, describes a cloud browser that can read pages, click buttons, fill forms, and run steps on supported public and signed-in sites. Your username and password go into that secure form, not into the large language model powering the chat.
OpenAI says the credentials entered there are not visible to the model, and ChatGPT does not store those sign-in details. Sign-in requests are also reviewed by an additional model for signs of phishing or deception. That second check is meant to catch deceptive login pages before you hand over access.
You can manage which sites ChatGPT may reach in Settings > Cloud browser, choosing Always ask, Auto approve, or Always allow. Some websites may restrict AI agents, so not every service will cooperate. That friction is part of the new reality: automation only works where sites allow it.
For readers who follow how digital habits age into new norms, this belongs with other Nostalgia: Then & Now stories about handing off chores we once did only in a browser tab. Yesterday’s DIY login marathon is today’s delegated cloud session—with confirmation gates when the stakes rise.
Why does this matter for everyday account safety?
The top question most people ask is simple: is my password going into the chatbot? Based on OpenAI’s description in the Mashable report, the intended answer is no—not into the model, and not into stored chat memory as a saved credential. That distinction matters more than the feature list.
It also matters because ChatGPT users can already connect bank accounts and medical records to the platform. Layering agent-style browsing on top of those links makes the “keys over” feeling real. The product is no longer only a writing partner; it is a task agent that can act on signed-in sites after you approve the path.
Availability is limited to Plus, Pro, and Business on mobile and web for now. If you are on a free plan, this is a capability to watch rather than use today. If you are eligible, treat first runs like any new autopilot: start with low-risk tasks, keep confirmation prompts on, and use Always ask until you trust the pattern.
Then versus now is stark. Then, finishing insurance paperwork or a flight change meant logging in yourself—or tempting fate by pasting secrets into a chat. Now, chatgpt can now things through a paused conversation, a remote browser, and a secure form designed so the model never sees the password. The buzz is real; the credibility hinges on whether users still treat confirmation screens as non-negotiable.
OpenAI framed ChatGPT Work as a place to gather those workflows in one spot. Mashable notes the tool aims beyond boardrooms into home utilities, travel, packages, and appointments. That breadth is why the story travels: it is not a niche plugin, but a claim that everyday account chores can run without exposing login details to the AI itself.
Keep expectations grounded. Supported sites, two-factor steps, and site-side blocks will shape what works week to week. Still, the core promise—act on your behalf without the model viewing your username and password—is the clearest answer to why this update matters in August 2026.