The gap between passing AI safety tests and knowing whether an AI system is safe Continue reading on Medium »