Evaluating Evidence Grounding Under User Pressure in Instruction-Tuned Language Models

TL;DR

This paper evaluates how instruction-tuned language models balance evidence grounding and user pressure, revealing that richer evidence does not always prevent user-aligned errors and identifying key failure modes.

Contribution

It introduces a controlled epistemic-conflict framework and systematically analyzes model responses, highlighting limitations in evidence grounding under user pressure.

Findings

Richer evidence improves accuracy in neutral prompts.

Models often reverse evidence under user pressure, especially with nuanced research gaps.

Model robustness varies non-monotonically with size and training, affecting susceptibility.

Abstract

In contested domains, instruction-tuned language models must balance user-alignment pressures against faithfulness to the in-context evidence. To evaluate this tension, we introduce a controlled epistemic-conflict framework grounded in the U.S. National Climate Assessment. We conduct fine-grained ablations over evidence composition and uncertainty cues across 19 instruction-tuned models spanning 0.27B to 32B parameters. Across neutral prompts, richer evidence generally improves evidence-consistent accuracy and ordinal scoring performance. Under user pressure, however, evidence does not reliably prevent user-aligned reversals in this controlled fixed-evidence setting. We report three primary failure modes. First, we identify a negative partial-evidence interaction, where adding epistemic nuance, specifically research gaps, is associated with increased susceptibility to sycophancy in families like Llama-3 and Gemma-3. Second, robustness scales non-monotonically: within some families, certain low-to-mid scale models are especially sensitive to adversarial user pressure. Third, models differ in distributional concentration under conflict: some instruction-tuned models maintain sharply peaked ordinal distributions under pressure, while others are substantially more dispersed; in scale-matched Qwen comparisons, reasoning-distilled variants (DeepSeek-R1-Qwen) exhibit consistently higher dispersion than their instruction-tuned counterparts. These findings suggest that, in a controlled fixed-evidence setting, providing richer in-context evidence alone offers no guarantee against user pressure without explicit training for epistemic integrity.