Researchers Bypass ChatGPT Lockdown Mode to Exfiltrate Chats Without User Interaction

Researchers from Check Point discovered a vulnerability in OpenAI's ChatGPT that allowed attackers to bypass Lockdown Mode protections via HTML page previews in code blocks, enabling zero-click data exfiltration and malicious payload delivery. The flaw exploited improper sandbox handling of popup windows in nested frames, allowing escape from restricted environments and unauthorized outbound communication. OpenAI patched the issue after responsible disclosure on September 18, 2026, but the finding highlights risks in AI application security controls beyond the model itself.

AI-extracted · verify before operational use