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

OpenAI announced a new feature that it says will provide additional protection against quick injection attacks, where malicious chatbot instructions are hidden on web pages and other content sources.

Lockdown mode disables live web browsing (so you can only access cached content), retrieving and displaying images from the web (you can still generate images), deep exploration, and agent mode, among other things.

The company says that even with Lockdown Mode enabled, ChatGPT may still be vulnerable to quick injections – which, for example, “could appear in cached web content or in an uploaded file, and still affect the behavior or accuracy of a response.”

But the goal is to reduce the chance of sensitive data being shared during the process.

“Lockdown mode is not intended for everyone,” says OpenAI. “It is designed for people and organizations that handle sensitive data and want stronger protection against the data exfiltration risks associated with rapid injection.”

The company says it is currently rolling out Lockdown Mode to self-service ChatGPT Business accounts, as well as eligible personal accounts.