A vulnerability in ChatGPT allowed attackers to circumvent the Lockdown Mode, a feature designed to prevent data leaks This article explores preview functionality chatgpt. . Check Point Research discovered that a hidden pop-up trick could access a user's complete chat history, create fake windows, and deliver malware.
This setting restricts web browsing to cached pages and disables live web access, image support in replies, Deep Research, Agent Mode, Canvas networking, live connectors, and file downloads.
A typical tactic involves tricking the AI into loading a link like attacker.example/collect?data= followed by private details. In February 2026, OpenAI integrated HTML preview functionality into ChatGPT code blocks, allowing users to preview pages directly within the chat. This feature uses two nested iframes, which are small, enclosed web pages.
Malicious prompts, custom GPT models, or shared chats could then send the entire conversation and uploaded files to the attacker's server.
Check Point reported the issue to OpenAI, which removed popup permissions from the outer preview iframe on September 18, 2026. No user action is necessary for this specific issue, but maintaining a few good habits can help mitigate similar risks. Always verify the address bar or sign-in page before entering a password, especially after a sudden logout.
Keep your desktop app updated, and refrain from pasting sensitive information, such as medical, financial, or work details, into any chat unless absolutely necessary.
