Researchers find way to bypass ChatGPT Lockdown Mode and steal chat data
- Severity
- High
- Sources
- 1
What happened
Check Point Research (Alexey Bukhteyev) found a flaw that bypasses ChatGPT Lockdown Mode, enabling silent exfiltration of full chat history, phishing via manipulated popups and malware delivery through embedded links. The issue is tied to the HTML page preview feature added to code blocks in February 2026, where the iframe sandbox permits allow-popups and allow-popups-to-escape-sandbox and the redefined window.open can be bypassed. In ChatGPT Classic the attack needs zero user interaction. The article is truncated and does not give a CVE, a fix status or affected versions; the earlier DNS tunnelling flaw in the code-execution runtime was reported as fixed.
For security analysts · Written by AI from 1 source · Wrong?
What to do
- Be cautious with shared chats, custom GPTs and prompts from unknown sources.
- Watch for an OpenAI fix for this bypass, as the article does not confirm one.
Written by AI from the sources, checked against them
Full story timeline · 1 update
Who reported · 1 outlet
Who and what is in it
Companies
Products