@zackwhittaker This reminds me of a friend who tests AI systems for a living. Her job is basically to see whether an AI can be tricked into doing things it wasn't supposed to do.
Is this the same general idea, or is it a completely different kind of vulnerability? 