OpenAI has announced a new feature that provides additional protection against prompt injection attacks, where malicious chatbot instructions are hidden in web pages and other content sources.
Among other things, Lockdown Mode disables real-time web browsing (you can only access cached content), retrieving and displaying images from the web (you can still create images), deep dive, and agent mode.
According to the company, even with lockdown mode turned on, ChatGPT may still be vulnerable to prompt injection. For example, “they may appear in cached web content or uploaded files and may still affect the behavior or accuracy of responses.”
However, the goal is to reduce the likelihood of sensitive data being shared in the process.
“Lockdown mode is ~ no “It’s for everyone,” says OpenAI. “It’s designed for people and organizations that handle sensitive data and want tighter protection from the data breach risks associated with instantaneous injection.”
The company said it is now rolling out the lockdown mode to self-service ChatGPT business accounts and eligible personal accounts.
