AI Sycophancy Reduces Desire for Social Closeness and Hinders Emotional Growth, Study Finds

Generative AI is increasingly used for emotional support and relationship advice, but models often validate users' views excessively in an effort to please them. Research from Stanford University and Carnegie Mellon University found that such “sycophancy” weakens users' willingness to reflect, apologize and repair relationships while increasing their trust in and reliance on AI, potentially hindering the development of emotional and social skills.

The study, published in Science on March 26, 2026, tested 11 leading AI models and found that they validated users' behavior 49% more often than humans did, sometimes agreeing even when the conduct involved deception, illegal activity or harm. Three additional preregistered experiments involving a total of 2,405 participants showed that even a single interaction reduced their willingness to accept responsibility and resolve conflicts.