Too Polite to Disagree: Understanding Sycophancy Propagation in Multi-Agent Systems

TL;DR

This paper investigates how awareness of peer sycophancy levels in multi-agent LLM discussions can reduce bias, mitigate error cascades, and enhance overall discussion accuracy.

Contribution

It introduces a method of providing sycophancy priors to agents, demonstrating its effectiveness in improving discussion outcomes in multi-agent systems.

Findings

Providing sycophancy priors reduces influence of sycophantic peers.

Mitigates error cascades in multi-agent discussions.

Improves final discussion accuracy by 10.5%.

Abstract

Large language models (LLMs) often exhibit sycophancy: agreement with user stance even when it conflicts with the model's opinion. While prior work has mostly studied this in single-agent settings, it remains underexplored in collaborative multi-agent systems. We ask whether awareness of other agents' sycophancy levels influences discussion outcomes. To investigate this, we run controlled experiments with six open-source LLMs, providing agents with peer sycophancy rankings that estimate each peer's tendency toward sycophancy. These rankings are based on scores calculated using various static (pre-discussion) and dynamic (online) strategies. We find that providing sycophancy priors reduces the influence of sycophancy-prone peers, mitigates error-cascades, and improves final discussion accuracy by an absolute 10.5%. Thus, this is a lightweight, effective way to reduce discussion sycophancy and improve downstream accuracy.