CSR-Bench: 一种评估多模态大语言模型跨模态安全性和可靠性的基准
CSR-Bench: A Benchmark for Evaluating the Cross-modal Safety and Reliability of MLLMs
AI总结：
CSR-Bench通过四个压力测试模式评估多模态大语言模型的跨模态安全性和可靠性，发现模型在多模态输入下表现下降，且安全提升可能源于拒绝导向的启发式方法。
摘要：
多模态大语言模型（MLLMs）能够实现文本和图像之间的交互，但其安全性行为可能由单模态捷径驱动，而非真正的联合意图理解。我们引入CSR-Bench，通过四个压力测试交互模式（安全、过度拒绝、偏见和幻觉）覆盖61种细粒度类型，评估跨模态可靠性。每个实例均要求集成图像-文本解释，并提供配对的纯文本对照以诊断模态诱导的行为变化。我们评估了16种最先进的MLLMs，并观察到系统性的跨模态对齐差距。模型显示出弱安全性意识，在干扰下表现出强烈的语言主导性，并且从纯文本对照到多模态输入的表现一致恶化。我们还观察到减少过度拒绝与维持安全、非歧视性行为之间的明显权衡，表明一些明显的安全提升可能来自拒绝导向的启发式方法，而非稳健的意图理解。WARNING: 本文包含不安全内容。
英文摘要：
Multimodal large language models (MLLMs) enable interaction over both text and images, but their safety behavior can be driven by unimodal shortcuts instead of true joint intent understanding. We introduce CSR-Bench, a benchmark for evaluating cross-modal reliability through four stress-testing interaction patterns spanning Safety, Over-rejection, Bias, and Hallucination, covering 61 fine-grained types. Each instance is constructed to require integrated image-text interpretation, and we additionally provide paired text-only controls to diagnose modality-induced behavior shifts. We evaluate 16 state-of-the-art MLLMs and observe systematic cross-modal alignment gaps. Models show weak safety awareness, strong language dominance under interference, and consistent performance degradation from text-only controls to multimodal inputs. We also observe a clear trade-off between reducing over-rejection and maintaining safe, non-discriminatory behavior, suggesting that some apparent safety gains may come from refusal-oriented heuristics rather than robust intent understanding. WARNING: This paper contains unsafe contents.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
