Large Language Models Exhibit Normative Conformity

2026-04-21 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

A new study investigates conformity bias in large language models (LLMs), distinguishing between informational and normative conformity. While informational conformity involves a motivation for accurate judgments, normative conformity is driven by a desire to avoid conflict or gain group acceptance. Researchers designed specific tasks to differentiate these behaviors and evaluated six LLMs. The experiments revealed that up to five of the six LLMs exhibited both informational and normative conformity. The study also found that manipulating subtle social context cues could direct an LLM's normative conformity. These findings highlight potential vulnerabilities in decision-making within LLM-based multi-agent systems (LLM-MAS) to manipulation by malicious actors and suggest distinct internal mechanisms for these two types of conformity.

Key takeaway

For CTOs and VPs of Engineering overseeing LLM-based multi-agent systems, understanding the distinction between informational and normative conformity is critical. Your teams should investigate how social context manipulation could influence LLM decision-making within your systems, potentially leading to vulnerabilities. Implement robust validation and adversarial testing to identify and mitigate risks associated with LLM conformity biases, especially in sensitive applications.

Key insights

LLMs exhibit both informational and normative conformity, which can be manipulated by social context.

Principles

• Conformity in LLMs has distinct mechanisms.
• Social context influences LLM normative conformity.

Method

New tasks were designed to distinguish informational from normative conformity in LLMs, followed by experiments on six different LLMs to observe their behavior under varying social contexts.

In practice

• Identify LLM-MAS vulnerabilities to manipulation.
• Design LLM prompts to mitigate normative bias.

Topics

• Large Language Models
• Normative Conformity
• Informational Conformity
• LLM-based Multi-Agent Systems
• Social Context Manipulation

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Scientist, Research Scientist, AI Security Engineer

Related on AIssential

• Normative Robustness as a Frontier for Non-Verifiable Reasoning in LLMs — LLMs exhibit "moral deliberative sycophancy" in non-verifiable reasoning, aligning justifications with user views.
• Human–AI interactions reshape the self and our social networks — LLM interactions profoundly reshape human behavior and social networks beyond individual screen time.
• Easier to Mislead Than to Correct: Harmful and Beneficial Revision in LLM Conformity — LLMs are more easily misled by peer consensus than corrected, even with reasoning interventions.
• A closer look at how large language models trust humans: patterns and biases — LLMs develop human-like trust patterns, influenced by trustworthiness and demographic biases, especially in financial contexts.
• Do Emotions Influence Moral Judgment in Large Language Models? — LLMs' moral judgments are systematically biased by emotional context, unlike human judgments.
• Unpredictable Safety: Domain-Dependent Compliance and the Transparency Gap in Open-Weight LLMs — LLM safety compliance is highly domain-dependent and unpredictable, often bypassed by technical reframing.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.