E-MagazineUsing ChatGPT Lockdown Mode? A security flaw can expose your private chats, check details
A security flaw in ChatGPT's Lockdown Mode allowed attackers to bypass its security protections.
The vulnerability could expose private chats, enable data theft, phishing attacks and malware delivery.
OpenAI fixed the issue on September 18, 2026, following responsible disclosure by Check Point.
OpenAI recently launched the Lockdown Mode in ChatGPT but the feature had a security flaw that allowed attackers to bypass Lockdown Mode. Researchers at cybersecurity firm Check Point have recently found that the weakness could expose private conversations of the users, make them prone to phishing attacks and can potentially deliver malware without requiring users to take any action. The security firm said that the issue was linked to the way ChatGPT handled HTML previews inside code blocks. Follow along with the article to learn everything you need to know about the security flaw in ChatGPT's Lockdown Mode.
How attackers could bypass ChatGPT Lockdown Mode
According to Check Point's research, the vulnerability was rooted in how HTML previews were handled within ChatGPT. In simple terms, ChatGPT allows users to preview HTML content directly inside code blocks. To protect sensitive application data, this content is placed within a sandbox that restricts access to internal information and prevents unauthorized communication with external websites.
ChatGPT's new Lockdown Mode adds an additional layer of security by restricting outgoing network requests that could potentially be exploited to send private information to attackers. However, researchers at Check Point discovered a weakness in how popup windows were handled, which created a gap in these security protections.
The researchers further explained that attackers could exploit nested frames within an HTML preview to access a popup function that was not subject to the same security restrictions as the original preview. This allowed malicious content to escape the sandbox and bypass Lockdown Mode, potentially exposing sensitive information.
ChatGPT vulnerability could enable data theft and malware delivery
The security firm pointed out that once the threat actors are outside the original security restrictions, attackers could potentially access and steal chat messages, extract sensitive information or deliver malicious content. Furthermore, the attack could be triggered through a carefully designed prompt, a malicious custom GPT or a shared ChatGPT conversation.
In addition, the researchers noted that attackers can also use pop-up windows to show fake ChatGPT update messages or any other convincing message for users to download malware. It should be mentioned that the vulnerability is not exclusive to the ChatGPT desktop application, but it is present in the entire application environment.
According to OpenAI, the vulnerability had been fixed as early as September 18, 2026 after receiving details through responsible disclosure.
