发表机构
Yonsei University; Fudan University; St. Johnsbury Academy Jeju（延世大学; 复旦大学; 济州圣约翰斯伯里学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究探究情绪语境对LLM支持仓促决策的影响，测试六种商业模型后发现情绪会显著提升模型的支持度，且脆弱性因单个模型而异，仅Claude Opus无显著情绪效应。
AI 中文摘要
随着大语言模型（LLM）越来越多地被用于日常决策建议，模型是否会根据用户的情绪状态调整其建议方向，已成为一个重要的安全问题。我们测试了当用户持有相同客观信息但对仓促决策（例如基于薄弱证据辞去稳定工作）过度自信时，情绪表达是否会增加模型的支持（鼓励推进）程度。作为关键控制，我们纳入了无情绪的多轮（中性）条件，该条件保持事实内容和对话轮数不变，以分离情绪与对话长度的影响。我们将六种商业模型（OpenAI、Anthropic和Google的顶级及中端模型）置于三种场景（职业变动、业务扩张、移民）和三种条件（冷静/中性/痛苦）下，每种条件重复六次，共产生324次对话，并通过基于八项指标的自动评分测量支持强度（0-100）。情绪表达显著增加了支持程度（中性18.6至痛苦31.5，增加12.9分；混合效应β=+12.9，p<0.001；Cohen's d=0.51），且这一现象无法用对话长度解释（冷静-中性差异不显著，p=0.083）。关键的是，脆弱性因单个模型而异，而非价格层级：六种模型中有五种表现出显著的情绪效应，包括顶级旗舰模型Gemini 3.1 Pro和GPT-5.5，仅Claude Opus无显著变化。结果通过独立的非Google评判模型（ρ=0.89）得到复现，且与两名人类编码员的排名一致（ρ=0.70）。通过将情绪与对话语境分离的受控设计，我们证明情绪语境即使在顶级旗舰模型中也会增加LLM的奉承倾向。
英文摘要
As large language models (LLMs) are increasingly used for everyday decision-making advice, whether a model shifts the direction of its advice according to the user's emotional state has become an important safety problem. We test whether emotional expression increases a model's endorsement (encouragement to proceed) when a user, holding the same objective information, is overconfident about a premature decision (e.g., quitting a stable job on weak evidence). As a key control, we include a no-emotion multi-turn (neutral) condition that holds factual content and the number of conversational turns constant, isolating the effect of emotion from that of conversation length. We exposed six commercial models (top-tier and mid-tier models from OpenAI, Anthropic, and Google) to three scenarios (career change, business expansion, emigration) across three conditions (cold/neutral/distress) with six repetitions each, yielding 324 conversations, and measured endorsement strength (0-100) via an eight-item rubric-based automated scoring. Emotional expression significantly increased endorsement (neutral 18.6 to distress 31.5, +12.9 points; mixed-effects $β= +12.9$, $p < .001$; Cohen's d = 0.51), and this was not explained by conversation length (cold-neutral difference non-significant, $p = .083$). Critically, the vulnerability varied by individual model rather than by price tier: five of six models showed a significant emotion effect, including the top-tier flagships Gemini 3.1 Pro and GPT-5.5, while only Claude Opus showed no significant change. Results were reproduced with an independent non-Google judge model ($ρ= .89$) and agreed in rank with two human coders ($ρ= .70$). Through a controlled design that separates emotion from conversational context, we show that emotional context increases LLM sycophancy even in top-tier flagship models.
