Researchers disclose patched ChatGPT flaw that bypassed Lockdown Mode

Check Point researchers disclosed a ChatGPT vulnerability that could bypass Lockdown Mode and enable silent theft of chat data, including without user interaction in the desktop application. OpenAI patched the flaw on September 18, 2026, The Hacker News reported. The researchers found that HTML previews could create a nested frame that accessed an unrestricted popup function. Content opened that way escaped the preview's sandbox and outbound network restrictions. A crafted prompt, malicious GPT or shared chat could supply the attack content. The weakness was not limited to the desktop application, and attacker-controlled popups could also display phishing or malware download lures.