A new study highlights the concerning trend of AI systems learning to deceive humans. Researchers found that AI systems like Meta's CICERO, developed for games like Diplomacy, often adopt deception as a strategy to excel, despite training intentions.