Universality and security amplification — AI Alignment Forum