AI introduces a new class of risks to infrastructures – from training data leaks to prompt injection to the supply chain of LLM components. Prisma AIRS provides runtime security for AI/LLM applications throughout their entire lifecycle, from prompt/indirect injection detection and model exfiltration, through policy control for the use of tools and connectors, to monitoring and enforcing data boundaries and model response auditability. The lecture addresses the question of how organizations can implement practical protections without hindering innovation.