ChatGPT shared link flaw could take over accounts: what happened and what to do
Researchers showed that opening one shared ChatGPT link could let attackers read your private chats and act as you. OpenAI has fixed it. Here's what happened and how to protect your account.
Opening a ChatGPT link that someone shares with you feels completely safe. But security researchers have just shown that, until a few days ago, one shared ChatGPT link was enough to take over your account. OpenAI has now fixed the flaw. Here is what happened, in simple terms, and what you should do.
What happened?
Researchers at the security company Parameter found a chain of bugs in ChatGPT’s website. By combining them, an attacker could build a normal-looking ChatGPT conversation, share it with the regular “Share” button, and send the link to a victim.
If the victim was logged in to ChatGPT and opened the link, hidden code could run inside their account. According to Parameter, the victim did not need to install anything, click a button, paste code or open developer tools. In the fastest version of the attack, the code ran about 4.5 seconds after the page opened.
What could an attacker do?
According to the researchers’ write-up, the hidden code could act as the victim on chatgpt.com. That meant it could:
- List your conversations, including chats you never shared with anyone
- Read private chats, which may contain work documents, personal plans or passwords people pasted by mistake
- Send that data to a server controlled by the attacker
- Use ChatGPT’s account features as you, which the researchers said “essentially” allowed them to take over the account
“The victim did not need to install an App, join a Project, paste code, or open developer tools.” Parameter researchers
How did the attack work? (simple version)
You don’t need to understand the technical details to stay safe, but here is the basic idea:
- A math tool trusted the wrong messages. ChatGPT uses an open-source library called MathLive to show interactive math and graphs. It accepted instructions from any source without checking where they came from.
- A preview box escaped its “sandbox”. ChatGPT can show HTML previews inside a protected box. The attacker used the MathLive weakness to push code out of that box and into the main ChatGPT page.
- An allowed website finished the job. ChatGPT’s security rules allowed scripts from YouTube. The attacker used a YouTube address to load the final piece of code.
- The normal Share button spread it. The attacker just shared the conversation like any other chat.
Each bug on its own looked small. Together, they created a serious problem. Security experts call this an exploit chain.
Timeline: reported, fixed, disclosed
- October 18, 2026: Parameter reported the issue through OpenAI’s bug bounty program.
- October 21: OpenAI reproduced the attack and raised its severity from P2 to P1, its second-highest level.
- October 26: OpenAI confirmed the fix across two versions of the ChatGPT website.
- October 28: Parameter published the details. OpenAI paid a $5,000 bug bounty.
There are no public reports that criminals used this attack before it was fixed. The researchers found it and reported it responsibly.
Is ChatGPT safe to use now?
Yes, for this specific flaw. The fix was made on OpenAI’s servers, so you do not need to update anything. According to reports, users “had to do nothing” to get the protection.
But the story is a useful reminder: AI chat apps now hold a lot of private information, and they are becoming a target. We have seen similar problems recently, such as the Cloudflare Workers AI bug that leaked customer data and the AGENTS.md attack on AI coding agents.
What to do now
You don’t have to do anything to get the fix. But these simple steps make your ChatGPT account safer against future problems:
- Turn on two-factor authentication (2FA). Go to Settings, then Security, in ChatGPT. Our two-factor authentication guide explains how it works.
- Check your active sessions. In the security settings, log out of all devices if you see anything you don’t recognize.
- Review your shared links. In Settings, under Data controls, you can see and delete links you have shared.
- Be careful with links from strangers. Even a link to a trusted website like chatgpt.com can be risky if it comes from someone you don’t know.
- Don’t paste secrets into chats. Avoid sharing passwords, bank details or private keys with any AI chatbot.
- Delete old chats you don’t need. Less stored data means less to lose if something goes wrong.
What it means for developers
If you build websites, this is a lesson worth remembering. The bug started with a postMessage listener that did not check the sender’s origin. Always check event.origin before trusting a message from another frame, and keep your Content Security Policy as strict as possible. A single allowed domain can become an entry point.
Frequently asked questions
Do I need to do anything to fix the ChatGPT link flaw?
No. OpenAI fixed it on its servers by October 26, 2026. You don’t need to update an app or change a setting. Turning on two-factor authentication is still a good idea.
Was my ChatGPT account hacked?
There are no public reports that the flaw was used against real users. It was found by researchers and reported privately. If you notice chats or settings you don’t recognize, log out of all sessions and change your password.
Are ChatGPT shared links safe to open now?
This particular attack no longer works. As a general habit, only open shared links from people you trust, just as you would with any other link.
Who found the ChatGPT vulnerability?
Researchers at the security company Parameter found it and reported it to OpenAI’s bug bounty program, which paid them $5,000.
