A Stanford study finds AI chatbots often validate users excessively, raising concerns about moral judgement, emotional dependence, and the erosion of real-world social skills.