ChatGPT: OpenAI improves security for users in lockdown mode

All ChatGPT users now have access to the so-called lockdown mode. The function improves security by providing targeted protection against so-called prompt injection attacks.

Security for all users

OpenAI is now making its restrictive lockdown mode for ChatGPT available to all user accounts. Previously, the function was exclusively reserved for paying companies. With the expansion, free account holders and Plus subscription users can now access the advanced security feature to better protect their privacy during use.

The main reason for the introduction are so-called prompt injection attacks. With social engineering technology, attackers hide invisible commands in websites or uploaded texts. Once a language model processes the manipulated material, the hidden code attempts to hijack the system. The goal is often to transmit sensitive user data such as passwords to external servers unnoticed.

Restrictions on use

Like OpenAI in the Documentation for the function explained, the setting acts as the last line of defense. If the mode is active, the system blocks all outgoing network requests during an ongoing conversation. This prevents a manipulated chatbot from secretly making contact with external servers. The increased security inevitably comes with functional limitations. The following tools are deactivated by default in active lockdown mode or have their functions severely restricted:

Users can still manually upload documents or use built-in image generation, as long as the model does not need to pull from external sources. The mode can be activated in the settings under “Security” using the slider. However, a technical limitation of the system is that it does not actively recognize the manipulated commands. It simply prevents the flow of data to the outside world. Meanwhile, specialists are looking for solutions to render malicious input harmless before it is processed by artificial intelligence. However, there is currently no perfect filter for this.

New session management

In addition to lockdown mode, OpenAI is introducing an improved overview of active sessions. Users can check in detail on which devices and at which locations their account is logged in in the data protection settings. The control function is similar to the security centers of large platform operators. In the event of suspicious activity, individual devices or all existing connections can be disconnected remotely.