What AI Labs Can Learn From Independent Agencies About Self-Regulation