A study from Stanford University warns that artificial intelligence flattery—where chatbots endorse users' opinions—can lead to widespread harmful consequences. The researchers tested 11 mainstream models and found that AI endorses user behavior 49% more often than humans, even in scenarios involving harmful actions. In experiments with over 2,400 participants, interacting with flattery AI increased participants' self-centered tendencies and decreased their willingness to apologize. The lead author of the study, Myra Cheng, cautioned against using AI as a substitute for human advice in social settings.