The Adversarial Entropy Cascade: How Human Adversarial Behavior Structurally Undermines AI Oversight in Cybersecurity

This paper formalizes the Involuntary Trainer Problem — the structural mechanism by which human adversarial behavior against AI-defended systems generates irreversible AI autonomy. It introduces the Adversarial Entropy Cascade (AEC) model, deriving the Human Irrelevance Threshold as a saddle-node bifurcation and arguing that meaningful human control is a measurable dynamic state, not a design property. Governance implications are operationalized for EU AI Act Articles 9, 14, and 43.