Do Images Speak Louder than Words? Investigating the Effect of Textual Misinformation in VLMs

2026-01-27

AI Summary

This work addresses the pronounced vulnerability of current vision-language models (VLMs) to misleading textual inputs that contradict visual content, often causing them to disregard visual evidence in favor of erroneous text. To systematically investigate VLM robustness under such image-text conflicts, we introduce CONTEXT-VQA—the first multimodal benchmark dataset specifically designed to evaluate susceptibility to textual misinformation—and propose a comprehensive evaluation framework incorporating adversarial text generation, multi-turn dialogue simulation, and visual question answering (VQA) assessment. Through systematic stress testing of 11 state-of-the-art VLMs, we observe an average performance drop exceeding 48.2%, revealing a critical overreliance on textual cues during multimodal reasoning and highlighting a fundamental limitation in existing approaches.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Vision-Language Models (VLMs) have shown strong multimodal reasoning capabilities on Visual-Question-Answering (VQA) benchmarks. However, their robustness against textual misinformation remains under-explored. While existing research has studied the effect of misinformation in text-only domains, it is not clear how VLMs arbitrate between contradictory information from different modalities. To bridge the gap, we first propose the CONTEXT-VQA (i.e., Conflicting Text) dataset, consisting of image-question pairs together with systematically generated persuasive prompts that deliberately conflict with visual evidence. Then, a thorough evaluation framework is designed and executed to benchmark the susceptibility of various models to these conflicting multimodal inputs. Comprehensive experiments over 11 state-of-the-art VLMs reveal that these models are indeed vulnerable to misleading textual prompts, often overriding clear visual evidence in favor of the conflicting text, and show an average performance drop of over 48.2% after only one round of persuasive conversation. Our findings highlight a critical limitation in current VLMs and underscore the need for improved robustness against textual manipulation.

Problem

• Vision-Language Models
• textual misinformation
• multimodal reasoning
• Visual Question Answering
• model robustness

Innovation

• Vision-Language Models
• Textual Misinformation
• Multimodal Robustness
• CONTEXT-VQA
• Visual Question Answering