A security vulnerability in ChatGPT’s shared conversation feature could have allowed attackers to execute JavaScript code within a victim’s authenticated session and access private chat histories.
Discovered by security researchers at Parameter, the flaw could be exploited through a specially crafted ChatGPT sharing link. In the most advanced version of the attack, simply opening the link was enough to trigger the exploit, without requiring any additional user interaction.
The researchers demonstrated that attackers could retrieve recent conversations, access private chats that had never been shared, and send extracted data to an external server.
The attack involved weaknesses in cross-frame messaging, the MathLive JavaScript library, and ChatGPT’s HTML Preview functionality, allowing malicious content to escape its intended security boundaries.
OpenAI classified the vulnerability as P1 after reproducing the issue and deployed a fix on October 26, 2026. The researchers reported no evidence of real-world exploitation, and users do not need to take any additional action.
Research: Parameter – Do the Math: How We Took Over ChatGPT Accounts