I want to understand what prompt injection means in AI and generative AI systems. How can malicious or unexpected inputs manipulate a model’s behavior or outputs? Can someone also explain the risks and common methods used to defend against prompt injection attacks?