Why Your ChatGPT Agent Now Only Needs Your Password Once | Edition 284

For as long as agentic AI tools have existed, there's been one wall they couldn't get past: the login screen. An agent could read a public page or draft inside an open document, but the moment a task needed your loyalty account, your work dashboard, or any site that simply wanted a password, it stalled and handed the job back to you.

OpenAI just said that wall is down for ChatGPT Work agents. And the real news isn't that a human can now log in on the agent's behalf — takeover-mode logins already existed. The real news is that the login sticks.

How the handoff actually works

When a running task hits a site that needs a login, the agent pauses and hands you control of the same cloud browser it's been driving. While you're in control, ChatGPT stops capturing screenshots of that window — specifically so a password or one-time code you type is never recorded or seen by the model, per OpenAI's own documentation. Once you finish signing in and hand control back, the agent picks the task back up exactly where it left off.

One real limit worth knowing: this doesn't turn every site into fair game. Sites can still block automated browsers outright, and any second factor you get sent — a text code, an authenticator app — still needs you to hand it over live during the takeover, same as the initial password.

What actually changed

Cookies now stick around across sessions the way they would in an ordinary browser, so a site you signed into on Monday is still signed in on Friday, for whatever task the agent is running by then. You can still sign out and clear cookies yourself from ChatGPT's data controls whenever you want a clean slate.

That one change — persistence, not the login handoff itself — is what turns a login wall from a per-task interruption into a one-time setup cost. Before this, handing an agent a recurring job on a site you sign into meant either logging back in every session, or not delegating that kind of task at all. Now the login is infrastructure you build once.

Major brands just took over a Times Square billboard, and every ad was built in Hightouch Ad Studio. Sweetgreen, HelloFresh, and Tripadvisor cleared the toughest room in advertising. Create, edit, and launch on-brand ads in a few clicks.

See what you can build

What that unlocks

Checking a loyalty balance. Actually completing a booking on a site that requires an account. Pulling a report out of an internal tool sitting behind your company's login. None of it requiring you to sit there re-authenticating on every single run — you do the sign-in once, and every future task on that site starts already logged in.

Where it ships

ChatGPT Work — the agent product this lives in, powered by OpenAI's new GPT-5.6 model family — launched July 9 and is rolling out to Pro, Enterprise, and Edu plans first, with Plus and Business close behind.

What this means for you

Set the login up once, deliberately: the next recurring task you're weighing whether to hand to a work agent, check whether it lives behind a sign-in you dread repeating — that's now the strongest case for delegating it, not a blocker. And the security model didn't change here, only the payoff did: your credentials were already protected during takeover; persistence just means you stop paying that setup cost every single session.

Office agents already cut real research work from two days to minutes in Edition 277. Persistent login is what lets that same agent keep working on the parts of your job that live behind a password — read the full edition, including the before/after breakdown, on the site.

Claude can be your analyst, editor, and strategist.

But most professionals are using it to fix grammar.

These 1,000+ Claude prompts take it from grammar tool to your most powerful AI work assistant.

Sign up for Superhuman AI and get:

Claim your free prompts