As AI agents reshape physical security operations, organisations face a dual challenge: a workforce unprepared for AI oversight, and prompt injection attacks, the digital equivalent of social engineering.  This tops the Open Worldwide Application Security Project (OWASP)'s list of threats for Large Language Models (LLMs). Cyrus shares how security leaders can build an AI-first threat posture, covering defensive prompt architectures, competency frameworks, and practical strategies to detect and counter AI-driven attacks.