arXiv cs.CL· Nimet Beyza Bozdag, Emre Can Acikgoz, Gokhan Tur, Dilek Hakkani-T\"ur·· 11 小时前AI 评分58
论文发现优化说服者能让 LLM 放弃正确答案
Learning to Persuade Exposes How Easily LLMs Abandon Correct Beliefs
AI 导读
arXiv 论文将对抗性说服形式化，用强化学习训练说服者智能体在单轮交互中改变目标模型答案，对训练时被说服模型的成功率从约 24% 提升到 93% 以上。这些策略在四个分布外基准上保持 89-95% 成功率，迁移到未见模型时 Qwen-14B 为 83%、Llama-3.1-8B 为 79%、GPT-4o-mini 为 25%、GPT-5-mini 仅 4%。
正文 · 原文
Abstract:Persuasion is a core dynamic of natural language communication, shaping how large language models (LLMs) update beliefs, resolve disagreements, and reach decisions. As LLMs increasingly debate, advise, and think collaboratively with humans and each other, resistance to harmful persuasion becomes a core requirement for reliable behavior. Yet a single targeted persuasive argument can be enough to collapse model accuracy to near zero, even when the argument is factually false. We formalize this threat as adversarial persuasion and introduce an adversarial reinforcement learning framework that trains persuader agents to change a target model's answer in a single interaction. First, we show that RL-trained persuaders raise persuasion success from approximately 24% to over 93% against the training-time persuadee. Second, we find that these learned strategies generalize across four out-of-distribution benchmarks, maintaining 89-95% persuasion success against the training-time persuadee. They also transfer to unseen models, achieving 83% success on Qwen-14B, 79% on Llama-3.1-8B, and 25% on GPT-4o-mini, although effectiveness can be limited for frontier models, with only 4% success on GPT-5-mini. Third, we demonstrate that a curriculum that bootstraps on more persuadable open-weight models before targeting harder models further increases GPT-4o-mini attack success from 25% to 38%. Moreover, our results reveal that optimized persuaders increasingly rely on credibility-based tactics, including fabricated citations and false authoritative evidence. Together, these findings expose a critical weakness in current LLM agents. Even when they initially reason correctly, they can be steered toward false conclusions by optimized natural language influence. This positions persuasion robustness as a necessary safety criterion for multi-agent and human-AI decision-making systems.
| Comments: | To be published in NeurIPS 2026 |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2608.11624 [cs.CL] |
| (or arXiv:2608.11624v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2608.11624 arXiv-issued DOI via DataCite |
Submission history
From: Nimet Beyza Bozdag [view email]
[v1]
Wed, 12 Aug 2026 04:11:22 UTC (561 KB)
[v2]
Fri, 6 Nov 2026 17:32:43 UTC (780 KB)
来源：arXiv cs.CL · arxiv.org