OpenAI has started rolling out Lockdown Mode, a new enhanced security feature for ChatGPT. It is designed to counter prompt injection attacks, where attackers embed malicious instructions in web pages, documents, or other content to manipulate the AI into performing unwanted actions.
OpenAI stresses that Lockdown Mode is not intended for everyone. It targets individuals and organizations handling sensitive data who want to minimize the risk of leaks. Essentially, it adds another security layer on top of ChatGPT's existing protections, including model and server infrastructure.
Once activated, Lockdown Mode limits certain ChatGPT capabilities. Users can still generate images and manually upload their own photos or documents, but the system cannot fetch images from the web, display inline pictures, or download files for analysis. Deep Research and Agent Mode are completely disabled, as these features frequently interact with external sources and network resources.
OpenAI clarifies that Lockdown Mode does not remove malicious instructions from content processed by ChatGPT. Instead, it restricts network requests and actions an attacker might use to extract sensitive account information. Settings for memory, file uploads, conversation sharing, and using chats to improve models remain independent and are not affected automatically.
To enable Lockdown Mode, go to ChatGPT settings under Safety and Security, then in the Advanced Security section, toggle Lockdown Mode on. If needed, you can temporarily disable protection for a specific chat via the status message above the dialogue window.
Alongside this, OpenAI is launching an active session manager. It shows the devices and browsers where your account is logged in and allows you to end individual sessions or sign out everywhere. If you suspect unauthorized access, OpenAI recommends changing your password, reviewing login methods, and contacting support.