Leading AI models are failing basic logic tests at alarming rates, and the consequences extend well beyond academic curiosity. New research shows that the same systems millions of people rely on for ...