Jailbreaking Multimodal Large Language Models using Multi-Clip Video
2026-06-01 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial IntelligenceComputation and Language
AI summaryⓘ
The authors studied how large language models that understand videos can be tricked into ignoring safety rules. They created a special video dataset with multiple short clips to test how different video features affect model safety. Their experiments showed that videos with more clips, movement, and varied scenes made the models more vulnerable than single images. They suggest using the relatively safer image mode as a way to improve defenses.
multimodal large language modelsvideo inputssafety alignmentjailbreak attacksMulti-Clip Video SafetyBenchdynamic vs static videocontext diversitymodel vulnerabilitydefense strategiesimage modality
Authors
Choongwon Kang, Seungjong Sun, Hyunmin Jun, Jang Hyun Kim
Abstract
As multimodal large language models (MLLMs) have advanced to process video inputs, concerns have emerged about their potential for malicious misuse. Prior jailbreak studies have shown that safety alignment in MLLMs can be bypassed through visual inputs, yet it remains unclear which properties of video inputs induce this vulnerability. To address this gap, we introduce Multi-Clip Video (MCV) SafetyBench, a dataset of 2,920 videos designed to evaluate how the diversity of video inputs affects the vulnerability of MLLMs. Each video consists of multiple short clips depicting diverse contexts related to a harmful query. Experiments on eight representative video MLLMs show that attack success consistently increases with the number of clips. Our results further indicate that the video modality is (1) more vulnerable than the image modality, (2) more vulnerable to dynamic videos than to static videos, and (3) more vulnerable when videos contain more diverse contexts. Building on these findings, we propose a defense strategy that leverages the relative robustness of the image modality.