Precursor checking for deceptive alignment — AI Alignment Forum