When an AI system's behavior deviates from its intended operational goals due to adversarial manipulation.