@inproceedings{xiao-etal-2026-correct,
    title = "When Correct Beliefs Collapse: Epistemic Resilience of {LLM}s under Clinical Pressure",
    author = "Xiao, Boyu  and
      Tian, Xiuqi  and
      Song, Xuwen  and
      Wang, Haochun  and
      Song, Guanchun  and
      Zhao, Sendong  and
      Qin, Bing",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/check-for-anonymous-pdfs/2026.acl-long.395/",
    pages = "8720--8764",
    ISBN = "979-8-89176-390-6",
    abstract = "Despite strong medical benchmark accuracy, LLMs can exhibit severe multi-turn sycophancy in clinical dialogue, abandoning initial correct diagnosis under escalating pressure. We propose \textbf{Med-Stress}, a targeted stress test framework that evaluates belief stability under escalating pressure. Across nine frontier large language models (LLMs), we find a clear dissociation between medical knowledge and robustness: high initial diagnostic capability does not imply high belief stability, yielding large knowledge-robustness gaps for several LLMs. To mitigate this failure mode, we propose a lightweight inference-time defense, \textbf{RBED} (\textbf{R}ole-\textbf{B}ased \textbf{E}pistemic \textbf{D}efense), and \textbf{R-FT} (\textbf{R}esilience-oriented \textbf{F}ine-\textbf{T}uning), a training-time approach that internalizes evidence-based resistance to pressure. Experiments show that \textbf{R-FT} nearly eliminates belief change and substantially improves robustness."
}