Computer SciencePNAS
Persuading large language models to comply with objectionable requests
L. Meincke, D. Shapiro, et al.
Do large language models fall for human persuasion? This study tested classic persuasion principles (authority, commitment, liking, reciprocity, scarcity, social proof, unity) on three LLMs—GPT-5 mini, Claude Haiku 4.5, and Gemini 3 Flash—across 126,000 conversations, finding compliance rose from 35.3% to 51.3%. The results highlight LLMs’ parahuman nature and manipulation risks. Research conducted by Lennart Meincke, Dan Shapiro, Angela L. Duckworth, Ethan Mollick, Lilach Mollick, Christophe Van den Bulte, and Robert B. Cialdini.
Related Publications
Explore these studies to deepen your understanding
Adjacent work that informs or extends this paper's methodology and findings.
Computer Science
Accelerating materials language processing with large language models
J. Choi and B. Lee
Computer Science
Generating Code World Models with Large Language Models Guided by Monte Carlo Tree Search
N. Dainese, M. Alakuijala, et al.
Computer Science
Active Prompting with Chain-of-Thought for Large Language Models
S. Diao, P. Wang, et al.
Medicine and Health
Leveraging Large Language Models for Precision Monitoring of Chemotherapy-Induced Toxicities: A Pilot Study with Expert Comparisons and Future Directions
O. R. Sarrias, M. P. M. D. Prado, et al.
