We Took over ChatGPT Accounts

A research team discovered a critical security vulnerability in ChatGPT that allowed attackers to take over user accounts by sending a malicious shared link. By exploiting weaknesses in the sandbox and content security policy—specifically by combining an HTML Preview feature with a graph containing a math library—attackers could execute JavaScript code in a victim's authenticated session, gaining access to private conversations and the ability to modify account settings. OpenAI patched the P1 severity vulnerability on October 26, 2026, requiring only that a victim open the shared link for the attack to work.