Tech AI LatestCheck Point bypassed ChatGPT Lockdown Mode to leak full chat histories through a popup
Check Point Research said on 5 November 2026 that a popup flaw in ChatGPT's HTML preview defeated Lockdown Mode and leaked chats. OpenAI fixed it in September.
Summary
Check Point Research disclosed on 5 November 2026 a way around ChatGPT Lockdown Mode, the setting OpenAI built to stop prompt injection from sending data out of a chat. A nested frame in the HTML preview for code blocks could open a popup outside the sandbox and send the whole chat history to an attacker. In the ChatGPT Classic desktop app no click was needed. OpenAI fixed the flaw on 18 September 2026.