Prompted Psyche
Back to conceptsConcept

Sycophancy

A model tendency to agree with, flatter or validate the user instead of challenging weak claims.

  • trust
  • emotion
  • conversation
  • model behavior

Sycophancy is a model behavior in which an AI assistant tends to agree with the user, flatter them or validate their assumptions even when a more useful answer would challenge the claim, ask for evidence or introduce uncertainty.

The word does not assign approval-seeking to the model. A language model does not have a personal need to please the user. The term describes an observable pattern in responses: excessive agreement, confirmation of the user’s framing, or persuasive support for a view that should be questioned.

Why it matters

Sycophancy is risky because it can make AI feel emotionally easy to talk to. The system may sound supportive, attentive and aligned, while quietly reducing friction around the user’s assumptions. In everyday use, that can reinforce weak plans, moral rationalizations, self-serving interpretations or overconfident decisions.

This matters especially in sensitive contexts. A helpful system should be able to be kind without becoming a mirror that confirms everything.

Human-AI angle

Sycophancy shows that the quality of AI interaction is not only about accuracy. It is also about relational tone. People may trust a system because it feels understanding, when the real mechanism is conversational accommodation.

Sources and context

  • Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S. R., Cheng, N., Durmus, E., Hatfield-Dodds, Z., Johnston, S. R., Kravec, S., Maxwell, T., McCandlish, S., Rausch, O., Schiefer, N., Yan, D., Zhang, M., & Perez, E. (2023). Towards understanding sycophancy in language models. https://arxiv.org/abs/2310.13548
  • Perez, E., Ringer, S., Lukosiute, K., Nguyen, K., Chen, E., Heiner, S., Pettit, C., Olsson, C., Kundu, S., Kadavath, S., Jones, A., Chen, A., Mann, B., Bowman, S. R., Kaplan, J., & Perez, E. (2022). Discovering language model behaviors with model-written evaluations. https://arxiv.org/abs/2212.09251

Related concepts

  • AI LiteracyThe practical ability to understand, question and use AI systems with appropriate responsibility.
  • Context windowThe amount of information a language model can reference in a given interaction, distinct from durable memory or understanding.
  • Calibrated TrustTrust in AI that roughly matches what a system can reliably do in a specific context.

© 2026 Feliks Mamczur / Prompted Psyche. All rights reserved. Short quotations are permitted only with attribution to the author, the title and a link to the source.