Self-regulation of safety in AI research — AI Alignment Forum