The Attack Chain Your AI System is Already Missing 

**Executive summary:** Organizations face a growing operational risk as adversaries embed generative AI and agentic systems into existing intrusion workflows; prompt filtering is brittle and insufficient because malicious intent often emerges from sequences of downstream actions rather than the initial prompt. The report urges defenders to adopt continuous behavioral monitoring, strict least-privilege defaults, robust identity and session controls, comprehensive logging of agent actions, and runtime guardrails to detect and contain agent misalignment before it becomes an operational threat.