Learn to attack and defend LLMs with prompt injection, RAG poisoning, and zero-trust agent security in 100 local labs.