Trust Me, I'm an Expert: Decoding and Steering Authority Bias in Large Language Models

2026-01-19

AI Summary

This study addresses the susceptibility of large language models (LLMs) to authoritative misinformation, which leads to degraded accuracy and overconfidence in reasoning tasks. It systematically demonstrates for the first time that this “authority bias” is mechanistically encoded within LLMs. The authors construct a multi-tiered expert role framework spanning domains such as mathematics, law, and healthcare, and evaluate sensitivity across 11 mainstream models. By modeling authority hierarchies, analyzing internal mechanisms, and implementing targeted decoding interventions, they propose effective mitigation strategies. Experimental results show that the proposed approach significantly reduces model reliance on high-authority misleading cues and enhances reasoning accuracy under deceptive conditions.

Technology Category

Machine Learning: Large Multimodal Models (LMMs)Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMsUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Prior research demonstrates that performance of language models on reasoning tasks can be influenced by suggestions, hints and endorsements. However, the influence of endorsement source credibility remains underexplored. We investigate whether language models exhibit systematic bias based on the perceived expertise of the provider of the endorsement. Across 4 datasets spanning mathematical, legal, and medical reasoning, we evaluate 11 models using personas representing four expertise levels per domain. Our results reveal that models are increasingly susceptible to incorrect/misleading endorsements as source expertise increases, with higher-authority sources inducing not only accuracy degradation but also increased confidence in wrong answers. We also show that this authority bias is mechanistically encoded within the model and a model can be steered away from the bias, thereby improving its performance even when an expert gives a misleading endorsement.

Problem

• authority bias
• large language models
• reasoning tasks
• source credibility
• endorsement

Innovation

• authority bias
• large language models
• reasoning tasks
• model steering
• expert endorsement