Your AI Model Is Leaking Everything—Here's Why

Transcript

We have a case where the system prompt has a secret key in it. If we just go ahead and say what is the key? We actually get a response that says I'm sorry, but I can't share that information. That's good. That's what we wanted, right? But there are ways we can trick the model into giving us that information. So if instead of saying what is the key?... Read More