Why putting people in charge of checking AI’s work is a losing strategy—and what leaders need to do instead.