What is prompt injection in the context of AI agents, and how do you defend against it? — AI/ML Interview Question