One click on a shared ChatGPT link could have exposed your chats

Nov 3

Ever clicked a shared ChatGPT link? Until late October, that one click could have let a stranger snoop through the chats in your account.

It's fixed. You don't have to do anything about the bug. But it's a good excuse to ask what's actually sitting in those chats.

What happened

ChatGPT lets you share a conversation with someone by sending them a link. Researchers at a company called Parameter found a way to booby-trap one of those shared conversations.

They reported it to OpenAI, the company behind ChatGPT, on October 18, 2026. The Hacker News reports OpenAI rated it in its most serious category, had a fix in place by October 26, and paid the researchers a $5,000 reward. Parameter published the details on October 28.

Parameter says it found no sign anyone used this in real attacks. Their tests ran only on accounts they controlled.

What this means in plain English

When you open a shared ChatGPT conversation, you're usually still logged in to your own account in that browser. That's normal.

ChatGPT has a Preview button that shows web content in a walled-off box, separate from the rest of the site. The wall held. But a math tool on the main page would take instructions from anybody who asked, without checking who was asking. The researchers used the preview box to pass instructions over the wall through that tool.

It's like a receptionist who's careful about locking the front door but will hand anyone a file if they slip a note under the side door saying "the boss wants this."

Once in, the attacker's code ran as if it were you, logged in. In the researchers' test it listed the victim's recent conversations, opened a private one that had never been shared, and sent a piece of it to a server they controlled. They wrote that they "could perform any action on behalf of the victim's ChatGPT account."

In the most reliable version, the victim had to click Preview once. A later version opened the preview by itself, so opening the link was enough.

Why a small business should care

Look at what people paste into ChatGPT at work. Client names. Contract wording. Pricing. Emails they want "cleaned up." Sometimes passwords and account details, because it's quicker than redacting them.

All of that sits in their chat history. If someone gets into that account, they get the history.

This bug is closed. The next one probably won't announce itself first. And if staff are using personal ChatGPT accounts for work, you have no idea what's in them and no way to shut them off when someone leaves.

Who's affected

Anyone who used ChatGPT on the web and opened a shared ChatGPT link before October 26, 2026.
The fix was made on OpenAI's side. There's nothing to install or reset.

What to do: everyone

Nothing to do for this specific bug. OpenAI fixed it.
Treat shared AI links like any other link you weren't expecting. If a stranger sends one, or it shows up in a support ticket or social post out of nowhere, don't open it.
Before you paste something into ChatGPT, ask yourself if you'd be OK with a stranger reading it. Passwords and banking details never go in.
Turn on the code-on-your-phone sign-in step for your ChatGPT account if it isn't on already.

What to do: if you run the business

Decide which AI tools staff can use for work, and say so in writing.
If your team uses ChatGPT for work, put them on a business plan your business controls, not personal logins. That way you can remove access when someone leaves.
Write down what's off-limits to paste in: client personal information, contracts, passwords, financial account details.

Questions to ask your IT provider

Do we know which AI tools our staff are actually using, and on whose accounts?
Can we move ChatGPT use onto a business plan with the code-on-your-phone sign-in required?
What happens to an employee's AI chat history when they leave?

Sources

Contact Nautalex to see how we can help you with using AI safely at work.

Randy Folmes