AI application security, prompt injection attacks, and defenses. This topic focuses on the security of LLM-based applications, especially prompt injection attacks. We will look at how an AI agent can be manipulated, for example by overriding its original instructions, leaking sensitive data, or changing how it performs a task. We will also study ways to defend against these attacks. This can include secure design patterns, such as limiting what the AI is allowed to do and designing the application so that the AI has only the permissions it really needs.