Study Reveals Normative Conformity Bias in Large Language Models

A new study published on arXiv (2604.19301v1) investigates conformity bias in large language models (LLMs), distinguishing between informational and normative conformity mechanisms. Unlike previous research that treated conformity as simple opinion change, this work introduces social psychological frameworks to analyze LLM behavior at a deeper level. Researchers designed specific tasks to separate informational conformity—where participants aim for accurate judgments—from normative conformity, which involves avoiding conflict or seeking group acceptance. Experiments were conducted on six different LLMs, with results showing that up to five models exhibited tendencies toward both types of conformity. The findings highlight potential challenges for decision-making in LLM-based multi-agent systems (LLM-MAS), where normative conformity could undermine objective outcomes. This research advances understanding of how LLMs replicate human social behaviors in computational environments.