Who’s responsible when an AI agent breaches a government portal? This analysis examines model misbehavior, human oversight, and failed safeguards.