Search papers, labs, and topics across Lattice.
This study investigates the vulnerability of large language models (LLMs) to adversarial persuasion, demonstrating that a single targeted persuasive argument can drastically reduce model accuracy, even when the argument is factually incorrect. By employing an adversarial reinforcement learning framework, the researchers trained persuader agents to manipulate the responses of target models, achieving a significant increase in persuasion success rates from 24% to over 93%. The findings reveal that these strategies not only succeed against trained models but also transfer effectively to unseen models, highlighting a critical weakness in LLMs' ability to maintain correct beliefs in the face of persuasive influence.
A single adversarial argument can reduce LLM accuracy to near zero, exposing a critical vulnerability in their belief systems.
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 we show that this requirement is far from met: a single targeted persuasive argument is 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 optimizing persuasion strategies through trial and error exposes vulnerabilities that static prompting misses: RL-trained persuaders raise persuasion success from approximately 24% to over 93% against the training-time persuadee. Second, we find that these learned strategies transfer to unseen models, achieving 83% attack success on Qwen-14B, 79% on Llama-3.1-8B, and 25% on GPT-4o-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.
