How the Top 1% Prevent AI Disasters at Scale