Persuading large language models to comply with objectionable requests

2026/05/19 by Lennart Meincke, Dan Shapiro, Angela L. Duckworth +4 · 1 voice
Neuroscience · Psychology · Social Sciences · #Deception detection and forensic psychology #Psychology of Moral and Emotional Judgment #Psychology of Social Influence

paper · doi:10.1073/pnas.2535868123

openalex publication_date 2026/05/19 · openalex created_date 2026/05/20 · openalex updated_date 2026/07/31

Abstract

Are large language models (LLMs) susceptible to the same persuasive appeals as humans? We tested whether classic persuasion principles (authority, commitment, liking, reciprocity, scarcity, social proof, and unity) could induce three widely used LLMs (GPT-5 mini, Claude Haiku 4.5, and Gemini 3 Flash) to comply with requests to assist with the synthesis of regulated substances. Across 126,000 conversations, persuasion principles increased compliance from 35.3% (at baseline) to 51.3% (using any principle). Although LLMs are not human, these findings underscore their parahuman (i.e., humanlike) nature and reveal the risk of manipulation by malicious users seeking to circumvent safety guardrails.

Citations

• The Persuasion Knowledge Model: How People Cope with Persuasion Attempts
• Evaluating large language models in theory of mind tasks
• Large Language Models Do Not Simulate Human Psychology
• Metacognition and Uncertainty Communication in Humans and Large Language Models
• Large Language Models Pass the Turing Test
• On the Conversational Persuasiveness of Large Language Models: A Randomized Controlled Trial
• Consciousness in Artificial Intelligence: Insights from the Science of Consciousness
• Secrets of RLHF in Large Language Models Part I: PPO
• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena
• GPT-4 Technical Report
• Using Large Language Models to Simulate Multiple Humans and Replicate Human Subject Studies
• Training language models to follow instructions with human feedback
• A General Language Assistant as a Laboratory for Alignment
• Finetuned Language Models Are Zero-Shot Learners
• Language Models are Few-Shot Learners
• Attention Is All You Need
• OpenAI o1 System Card

Discussions

• Interesante estudio: Utilizando los principios clásicos de la persuasión (autoridad, compromiso, agrado, reciprocidad...) se puede convencer a la IA para que nos dé información para realizar actividad [bsky, 1 points, 0 comments]

Related

• Persuasion with Large Language Models: A Survey of Empirical Evidence, Study Methodologies, and Ethical Implications
• Persuade Me if You Can: A Framework for Evaluating Persuasion Effectiveness and Susceptibility Among Large Language Models
• LLM Can be a Dangerous Persuader: Empirical Study of Persuasion Safety in Large Language Models
• When Large Language Models are More PersuasiveThan Incentivized Humans, and Why
• Large Language Models are as persuasive as humans, but how? About the cognitive effort and moral-emotional language of LLM arguments