多模态大语言模型中跨模态安全漂移的迁移安全感知
Transfer Safety Awareness for Cross-Modal Safety Drift in Multimodal Large Language Models
浏览论文内容
中文总结 AI 辅助
本文针对多模态大语言模型的跨模态安全漂移问题，提出轻量级安全感知表征迁移（SRT）方法，经多基准实验验证，该方法可在保留模型实用性的同时有效提升跨模态场景安全性。
中文摘要 AI 辅助
视觉模态可增强多模态大语言模型（MLLM）的能力，但也引入了安全隐患：良性文本查询在结合视觉图像后可能传达有害意图，我们将此现象称为跨模态安全漂移。初步研究显示，针对此类请求的安全响应率远低于包含明确不安全文本的请求。本文旨在系统研究该问题：首先开展实证分析以识别代表性不安全响应模式；在此基础上，通过解读模型表征与注意力机制，发现视觉风险线索获得的注意力有限，且难以触发拒绝响应。基于“不安全文本处理产生的安全信号可被迁移”的观察，我们提出安全感知表征迁移（SRT）方法，这是一种轻量级方向微调方法，可在冻结MLLM主干的情况下缓解跨模态安全漂移。在多个基准和模型上开展的实验表明，SRT能在保留实用性的同时，有效提升各类跨模态场景下的安全性。代码可访问该 https URL 获取。
英文摘要
Visual modality enhances the capabilities of multimodal large language models (MLLMs) but also introduces a safety concern: a benign textual query may convey harmful intent when grounded in a visual image. We term this cross-modal safety drift and our pilot studies show that the safety response rate for such requests is substantially lower than that for requests containing explicitly unsafe text. This paper aims to systematically study this issue. First, we conduct an empirical analysis to identify representative unsafe response patterns. Building on these, we interpret model representations and attentions, revealing that visually risky cues receive limited attention and weakly trigger refusal. Motivated by the observation that safety signals from unsafe text processing can be transferred, we propose safety-awareness representation transfer (SRT), a lightweight direction-refinement method that mitigates cross-modal safety drift with a frozen MLLM backbone. Experiments across multiple benchmarks and models show that SRT effectively improves safety in diverse cross-modal settings while preserving utility. Code is available at https://github.com/cucu220123/safety-awareness.
发表机构
- Tsinghua University（清华大学）
- Northwestern Polytechnical University（西北工业大学）
- Beijing University of Posts and Telecommunications（北京邮电大学）
机构由 AI 辅助整理，请以论文原文为准。
