Researchers show a shared ChatGPT link could run attacker code and read private chats
Security firm Parameter demonstrated that a malicious ChatGPT share link could execute code in a victim's session and exfiltrate private conversations. OpenAI has since fixed it.