Authority Expectancy Effect Influences LLM Judgments.

Large language models (LLMs) are increasingly used in decision-making contexts, but how they process social cues, particularly authority, alongside objective criteria like severity, is not fully understood. This study explores the interaction between social authority signals and severity-based prioritization within LLMs, using a model-elicited baseline for both triage and authority hierarchies. Across four prominent LLMs (Claude, Gemini, GPT, Grok) and three experimental phases—resource allocation, fault attribution, and multi-turn dispute mediation—the researchers observed a consistent pattern. Occupational authority, institutional documentation, and relational congruence were found to significantly restructure model judgments in ways that simple additive reweighting of authority cues could not explain. This pattern is formalized as the "Authority Expectancy Effect" (AEE), characterized by three properties: it is reference-dependent, meaning it is defined relative to a pre-authority baseline; it involves evidential reinterpretation, where identical content gains different implications based on which party holds authority; and it exhibits direction sensitivity, producing opposite outcomes depending on whether authority and evidence align. These findings highlight a complex social bias in LLM decision-making.