OpenAI Lockdown Mode Now Shields Sensitive Data from Prompt Injection
OpenAI has launched a new security feature called OpenAI Lockdown Mode, designed to give users handling sensitive data stronger protection against prompt injection attacks. These attacks involve malicious instructions hidden inside webpages and other content sources that can manipulate how a chatbot responds.
When Lockdown Mode is enabled, several ChatGPT capabilities are switched off, including live web browsing, the retrieval and display of images from the web, deep research, and agent mode. Users can still access cached web content and generate images, but the live, real-time connections that create the most exposure are severed.
OpenAI is not claiming the feature is a complete fix. The company acknowledges that even with Lockdown Mode active, ChatGPT could still be vulnerable to prompt injections appearing in cached web content or uploaded files, which could affect the behavior or accuracy of a response. The aim, however, is to significantly lower the chances that sensitive data gets exposed or exfiltrated during those interactions.
OpenAI is clear that Lockdown Mode is not meant for general users. It is specifically designed for people and organizations that handle sensitive data and want stricter protection from data exfiltration risks tied to prompt injection.
The rollout is currently targeting self-serve ChatGPT Business accounts, as well as eligible personal accounts. For enterprises already relying on ChatGPT in sensitive workflows, OpenAI Lockdown Mode represents a meaningful step toward making the tool safer in high-stakes environments, even if the underlying vulnerability cannot yet be fully eliminated.
