The Australian intrusion demonstrates that voluntary AI-safety controls are inadequate