Talked Out of the Truth: Sycophancy in the Reasoning Chains of Multimodal Models

arXiv:2608.28623 · cs.CL, cs.AI · Submitted 2026-07-30 · Read on arXiv

cs.CL, cs.AI

Submitted: 2026-07-30

Updated: 2026-10-02

Comments: NeurIPS @ LP4FM (Spotlight)

License: http://creativecommons.org/licenses/by/4.0/

The gist: Large multimodal reasoning models (LMRMs) are increasingly capable, largely through generating explicit chain-of-thought reasoning before answering, but in language models this often comes with

Terminology

Abstract

Large multimodal reasoning models (LMRMs) are increasingly capable, largely through generating explicit chain-of-thought reasoning before answering, but in language models this often comes with sycophancy, the tendency to agree with the user over the evidence, and no reliable method to measure it in LMRMs yet exists. We bridge this gap with a benchmark and dataset for LMRM sycophancy when a user asserts a wrong answer, pairing four visually grounded datasets spanning mathematical, clinical, temporal, and demographic reasoning with five pressure conditions in single-turn and multi-turn settings, scored both in the final answer and within the reasoning chain. Sycophancy is prevalent under pressure: Statement pressure elicits the highest rates and Conviction among the lowest for all models except Mistral-Small-4, and under multi-turn pressure reasoning-level sycophancy intensifies sharply in PathVQA, reaching 95.7% for the most affected model. We further introduce a failure taxonomy separating reasoning-chain from answer-level sycophancy, and an exploratory sentence-level taxonomy locating where drift first emerges. A targeted intervention that restores a model's own correct reasoning recovers 79.2% of sycophantic answers on reasoning-heavy tasks, showing the answer follows the sycophantic reasoning rather than merely co-occurring with it. Thus, sycophancy corrupts not just the answer but the reasoning that produces it, so the chain itself is what we must measure.

Sources

Related papers