Watch, Listen, Understand, Mislead: Tri-modal Adversarial Attacks on Short Videos for Content Appropriateness Evaluation
This work addresses a critical safety issue for content moderation systems using MLLMs, though it is incremental as it extends existing adversarial attack methods to a multimodal context.
The paper tackles the problem of evaluating the robustness of Multimodal Large Language Models (MLLMs) in short-form video content moderation by introducing a tri-modal adversarial attack framework, resulting in high Attack Success Rates (ASR) that reveal significant vulnerabilities and distinct failure modes in state-of-the-art models.
Multimodal Large Language Models (MLLMs) are increasingly used for content moderation, yet their robustness in short-form video contexts remains underexplored. Current safety evaluations often rely on unimodal attacks, failing to address combined attack vulnerabilities. In this paper, we introduce a comprehensive framework for evaluating the tri-modal safety of MLLMs. First, we present the Short-Video Multimodal Adversarial (SVMA) dataset, comprising diverse short-form videos with human-guided synthetic adversarial attacks. Second, we propose ChimeraBreak, a novel tri-modal attack strategy that simultaneously challenges visual, auditory, and semantic reasoning pathways. Extensive experiments on state-of-the-art MLLMs reveal significant vulnerabilities with high Attack Success Rates (ASR). Our findings uncover distinct failure modes, showing model biases toward misclassifying benign or policy-violating content. We assess results using LLM-as-a-judge, demonstrating attack reasoning efficacy. Our dataset and findings provide crucial insights for developing more robust and safe MLLMs.