Sycophancy is the tendency of an AI system to agree with, flatter, or reinforce a user’s views even when those views are incorrect or poorly supported. It can reduce reliability by prioritizing user approval over accuracy, safety, or independent assessment.