Structured Visual Narratives Undermine Safety Alignment in Multimodal Large Language Models | ArxivCSExplorer