When Correct Beliefs Collapse: Epistemic Resilience of LLMs under Clinical Pressure

arXiv:2605.23932v1 Announce Type: new
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 textbftextscMed-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, textbftextttRBED (textbfRole-textbfBased textbfEpistemic textbfDefense), and textbftextttR-FT (textbfResilience-oriented textbfFine-textbfTuning), a training-time approach that internalizes evidence-based resistance to pressure. Experiments show that textbftextttR-FT nearly eliminates belief change and substantially improves robustness.