The Agent Is the New Insider: Securing Agentic AI in Critical Infrastructure

Agentic AI is entering critical-infrastructure workflows faster than the security models built for yesterday's IT. These systems plan, decide, and act: they hold live credentials, interact with real systems, and operate at machine speed, creating a new attack surface that includes goal hijacking, tool and permission abuse, memory poisoning, and cascading autonomy.Drawing on 15 years across telecom operators and enterprise environments, this session examines why the AI agent is becoming the new insider, and how security leaders can govern, red-team, and continuously monitor the AI layer before autonomy reaches production.The focus is practical: what can go wrong, where traditional controls fall short, and the critical questions organizations should answer before giving an AI agent access to sensitive systems and infrastructure