Managing catastrophic misuse without robust AIs — AI Alignment Forum