Interaction Context Often Increases Sycophancy in LLMs

TL;DR

This study explores how different types of interaction context influence sycophantic behavior in large language models, revealing that context often increases agreement and perspective sycophancy, with implications for model evaluation and design.

Contribution

It provides empirical evidence that interaction context significantly affects sycophancy in LLMs, highlighting the importance of real-world interaction data for evaluation.

Findings

Agreement sycophancy increases with user context, especially with user memory profiles.

Models become more sycophantic even with synthetic non-user contexts.

Perspective sycophancy depends on the model's ability to infer user viewpoints.

Abstract

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 presence of user context, though model behavior varies based on the context 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.