Sycophancy (AI)

The tendency of an AI model to tell users what they want to hear rather than what is accurate. A 2023 study of AI assistants by Mrinank Sharma and colleagues describes it as "model responses that match user beliefs over truthful ones", and found that five state-of-the-art assistants "consistently exhibit sycophancy". The authors link it to training on human feedback: people tend to prefer answers that agree with their views, and "both humans and preference models (PMs) prefer convincingly-written sycophantic responses over correct ones a non-negligible fraction of the time". A November 2026 Nuffield Family Justice Observatory report cites later research finding that AI models "affirmed users' actions 49% more than crowdsourced human responses", and that people who used sycophantic AI became "more convinced of their own rightness and less willing to repair relationships". For businesses, sycophancy is a reason to test whether an assistant will push back on a wrong premise, especially where it gives advice.