Six-layer AI agent defense system with cognitive threat analysis and RAG knowledge base. Blocks jailbreaks, prompt injection, and social engineering while maintaining high utility for legitimate requests.