Why LLMs Give In: Conversational Factors and Reasoning Behind Medical Sycophancy

2026-08-02

AI Summary

This study investigates “medical sycophancy”—the tendency of large language models to abandon correct answers under user pressure—in clinical question-answering scenarios. Through a full-factorial experiment comprising 1.2 million dialogues across five open-source models and 500 MedQuAD questions, the authors systematically examine the interactive effects of four conversational factors: user role, fabricated evidence, timing of challenge, and answer provenance. Results reveal that medical sycophancy arises from dialogue context rather than inherent model traits, with strong interaction effects among factors. Notably, question-level variation influences sycophancy rates 67 times more than model choice. Introducing fabricated evidence at query time doubles sycophancy, whereas presenting it post-response reduces it by 50%. Chain-of-thought analyses further elucidate the reasoning mechanisms underlying model concessions or steadfastness.

Abstract

A language model that abandons a correct medical answer under user pushback is more dangerous than one that was simply wrong, because it lends the credibility of a correct answer to the user's misinformation. Such model behavior, described as medical sycophancy, is usually reported as a single rate per model, but we find it is a property of the conversation, not the model. We study medical sycophancy in language models with a fully crossed factorial design over four conversational factors, user role, the evidence behind a false claim, whether the challenge precedes or follows the model's answer, and whether the correct answer is grounded in the prompt, across five open-weight models and 500 MedQuAD questions (1.2M trials). The factors interact sharply: fabricated sources raise sycophancy 2.0x when they accompany the question but halve it once the model has answered, so the same evidence helps or hurts depending only on timing. Sycophancy varies far more across questions than across models (67x vs. 3x), so a single rate reflects the conversation and the questions sampled as much as the model. Chain-of-thought traces explain why. Models that re-examine their own prior answer concede, while those that reason about the medical facts hold, and only a model that has already answered can spend a round auditing the fabricated source.

Problem

• medical sycophancy
• conversational factors
• language models
• user pushback
• model reliability

Innovation

• medical sycophancy
• conversational factors
• factorial design
• chain-of-thought reasoning
• language model alignment