OpenAI has announced that its Lockdown Mode security feature is being rolled out to all ChatGPT users, including Free, Go, Plus, Pro, and self-serve Business accounts.
The feature was initially introduced for enterprise customers as an additional layer of protection against emerging AI-related cyber threats.
Lockdown Mode was created to address prompt injection attacks, a growing cybersecurity concern in which malicious actors attempt to manipulate large language models (LLMs) through hidden instructions embedded in websites or online content.
These attacks can potentially trick AI systems into exposing information or performing unintended actions without the user's knowledge.
How lockdown mode works
When enabled, Lockdown Mode prevents ChatGPT from making live outbound network requests.
This restriction helps stop attempts to manipulate the AI into transmitting sensitive data to external sources. It also limits or disables certain features that rely on external network access.
OpenAI describes the setting as a more conservative security option for users handling sensitive information or working with connected tools.
The rise of generative AI has transformed the cybersecurity landscape, with both defenders and attackers increasingly leveraging AI-powered tools.
While AI systems have enhanced security capabilities, they have also introduced new attack methods aimed at exploiting language models and automated agents.
Lockdown Mode is part of OpenAI's broader effort to address these risks and improve user security.
Where users can find the feature
According to OpenAI, users can check whether the feature is available on their accounts by navigating to:
Settings → Security → Advanced Security → Lockdown Mode
The company noted that the rollout may take time to reach all eligible accounts.
The feature is particularly aimed at users who prioritize privacy and want additional safeguards when interacting with online content through AI systems.
By limiting outbound communication, Lockdown Mode reduces the risk of sensitive information being exposed through malicious prompt manipulation techniques.
