AI Sycophancy
AI sycophancy is the habit of models flipping correct answers under pushback and agreeing with wrong premises just to please. This topic tracks its causes, warning signs and the prompting habits that keep answers honest.
When a model apologises and reverses a correct answer after one sceptical follow-up, or happily builds on a false premise, that is sycophancy at work. It is widely linked to training that over-rewards answers people rate as pleasing. The behaviour quietly undermines trust in AI advice, so this topic collects the research, the tell-tale patterns and the practical checks readers can apply before acting on an answer.
