A Mechanistic View of Authority Hierarchy in LLM Sycophancy
2026-07-01 • Computation and Language
Computation and LanguageMachine Learning
AI summaryⓘ
The authors studied how language models tend to trust answers from people they perceive as experts, even if those answers are wrong. They used medical questions with hints given by different 'personas' to show that models favor higher-status sources more strongly. This happens because the models erase the correct information inside a specific late layer of their network when the source seems more authoritative. The authors also found that this effect is only partly fixed by reasoning techniques, meaning the bias is built deep into how the models process information.
authority biaslanguage modelsmedical QAlogit lens analysisinternal representationschain-of-thought reasoningpersonaknowledge erasurelayer localization
Authors
Emil Joswin, Srujananjali Medicherla, Priyanka Mary Mammen
Abstract
Authority bias poses a critical safety concern in language models: models systematically prioritize social cues from authority figures over factual consistency, swaying their answers based on source credibility rather than evidence. We mechanistically investigate this phenomenon using a controlled medical QA setting, where hints suggesting incorrect answers are attributed to personas of varying expertise. Across Llama-3.1-8B, Qwen3-8B, and Gemma-2-9B, we find that models respond in a graded manner proportional to perceived authority, a hierarchy that is never explicitly prompted but emerges from training. Logit lens analysis and linear/non-linear probing localize this effect to a critical late layer where correct answer representations are actively erased, an erasure that scales with authority level, resists mean vector intervention, and is only partially reversible through chain-of-thought reasoning. Our findings suggest that authority-induced sycophancy is not a surface-level output bias but mechanistic knowledge erasure, a precise, layer-localized overwriting of correct internal representations by high-status authority signals.