Researchers Show ChatGPT Lockdown Mode Could Be Bypassed to Steal Chats
OpenAI has patched the flaw, but Check Point says the underlying weakness is not limited to the desktop app.
Check Point researchers disclosed a ChatGPT flaw that let attackers get around Lockdown Mode, the optional setting OpenAI introduced in June 2026 to curb data theft through prompt injection. The weakness sat in how HTML previews inside code blocks handled popup windows: nested sandboxed frames still allowed popups to break out of the sandbox, letting content open outside its restrictions. That opened a route for silent chat theft, phishing and malware delivery, including zero-click exfiltration in the ChatGPT desktop app. OpenAI patched the issue on September 18, 2026. The researchers said the same problem is not confined to the Electron desktop environment and argued that AI security controls are only as strong as the application around them.
Why it matters
The flaw shows that a safety setting is only as strong as the application around it. Users who turned on Lockdown Mode to guard against prompt injection could have had chat data taken without any action on their part, including in the desktop app. OpenAI's September 18, 2026 fix closes this particular route, but the researchers' warning that the same problem extends beyond the Electron environment means users cannot treat an optional protection as a complete guarantee.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.