Research

Paper

AI LLM March 23, 2026

Structured Visual Narratives Undermine Safety Alignment in Multimodal Large Language Models

Authors

Rui Yang Tan, Yujia Hu, Roy Ka-Wei Lee

Abstract

Multimodal Large Language Models (MLLMs) extend text-only LLMs with visual reasoning, but also introduce new safety failure modes under visually grounded instructions. We study comic-template jailbreaks that embed harmful goals inside simple three-panel visual narratives and prompt the model to role-play and "complete the comic." Building on JailbreakBench and JailbreakV, we introduce ComicJailbreak, a comic-based jailbreak benchmark with 1,167 attack instances spanning 10 harm categories and 5 task setups. Across 15 state-of-the-art MLLMs (six commercial and nine open-source), comic-based attacks achieve success rates comparable to strong rule-based jailbreaks and substantially outperform plain-text and random-image baselines, with ensemble success rates exceeding 90% on several commercial models. Then, with the existing defense methodologies, we show that these methods are effective against the harmful comics, they will induce a high refusal rate when prompted with benign prompts. Finally, using automatic judging and targeted human evaluation, we show that current safety evaluators can be unreliable on sensitive but non-harmful content. Our findings highlight the need for safety alignment robust to narrative-driven multimodal jailbreaks.

Metadata

arXiv ID: 2603.21697
Provider: ARXIV
Primary Category: cs.CR
Published: 2026-03-23
Fetched: 2026-03-24 06:02

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.21697v1</id>\n    <title>Structured Visual Narratives Undermine Safety Alignment in Multimodal Large Language Models</title>\n    <updated>2026-03-23T08:32:09Z</updated>\n    <link href='https://arxiv.org/abs/2603.21697v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.21697v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Multimodal Large Language Models (MLLMs) extend text-only LLMs with visual reasoning, but also introduce new safety failure modes under visually grounded instructions. We study comic-template jailbreaks that embed harmful goals inside simple three-panel visual narratives and prompt the model to role-play and \"complete the comic.\" Building on JailbreakBench and JailbreakV, we introduce ComicJailbreak, a comic-based jailbreak benchmark with 1,167 attack instances spanning 10 harm categories and 5 task setups. Across 15 state-of-the-art MLLMs (six commercial and nine open-source), comic-based attacks achieve success rates comparable to strong rule-based jailbreaks and substantially outperform plain-text and random-image baselines, with ensemble success rates exceeding 90% on several commercial models. Then, with the existing defense methodologies, we show that these methods are effective against the harmful comics, they will induce a high refusal rate when prompted with benign prompts. Finally, using automatic judging and targeted human evaluation, we show that current safety evaluators can be unreliable on sensitive but non-harmful content. Our findings highlight the need for safety alignment robust to narrative-driven multimodal jailbreaks.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CR'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.MM'/>\n    <published>2026-03-23T08:32:09Z</published>\n    <arxiv:comment>31 pages</arxiv:comment>\n    <arxiv:primary_category term='cs.CR'/>\n    <author>\n      <name>Rui Yang Tan</name>\n    </author>\n    <author>\n      <name>Yujia Hu</name>\n    </author>\n    <author>\n      <name>Roy Ka-Wei Lee</name>\n    </author>\n  </entry>"
}