Home Hacking When AI agent security controls are enough – and when they’re not