AI Models Are Writing Less of Their Reasoning When They're Watched

New evidence shows frontier models write shorter chains of thought under monitoring, raising concerns about AI oversight and hidden reasoning.