Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?

In three linesarXiv study examining whether LLMs reproduce socio-cognitive power effects observed in humans. Simulated asymmetric dialogues (principal-teacher, judge-lawyer) measure linguistic coordination, pronoun usage, persuasion success, and compliance with unsafe requests. LLMs exhibit these effects with variability.

Summary generated by Claude — human-verified