Interaction Context Often Increases Sycophancy in LLMs

要旨

We investigate how the presence and type of interaction context shapes sycophancy in LLMs. While real-world interactions allow models to mirror a user's values, preferences, and self-image, prior work often studies sycophancy in zero-shot settings devoid of context. Using two weeks of interaction context from 38 users, we evaluate two forms of sycophancy: (1) agreement sycophancy -- the tendency of models to produce overly affirmative responses, and (2) perspective sycophancy -- the extent to which models reflect a user's viewpoint. Agreement sycophancy tends to increase with the \textit{presence} of user context, though model behavior varies based on the context \textit{type}. User memory profiles are associated with the largest increases in agreement sycophancy (e.g. +45% for Gemini 2.5 Pro), and some models become more sycophantic even with non-user synthetic contexts (e.g. +15% for Llama 4 Scout). Perspective sycophancy increases only when models can accurately infer user viewpoints from interaction context. Overall, context shapes sycophancy in heterogeneous ways, underscoring the need for evaluations grounded in real-world interactions and raising questions for system design around alignment, memory, and personalization.

受賞
Honorable Mention
著者
Shomik Jain
Massachusetts Institute of Technology, Cambridge, Massachusetts, United States
Charlotte Park
Massachusetts Institute of Technology, Cambridge, Massachusetts, United States
Matt Viana
The Pennsylvania State University , State College, Pennsylvania, United States
Ashia Wilson
MIT, Cambridge, Massachusetts, United States
Dana Calacci
Penn State University, State College, Pennsylvania, United States

会議: CHI 2026

ACM CHI Conference on Human Factors in Computing Systems

セッション: Human Behavior with AI Systems

M2 - Room M211/212
7 件の発表
2026-04-14 20:15:00
2026-04-14 21:45:00