Systemic Misalignment: Exposing Failures of Surface-Level AI Alignment Methods