Minimizing Empowerment for Safety — AI Alignment Forum