研究揭示视觉语言模型场景状态安全漏洞,提出 SceneSafetyBench 与 S3TA 攻击法
Reading Scene States: Uncovering Safety Failures in Vision-Language Models
这篇论文把一个很隐蔽的漏洞讲清楚了:不改指令、只摆图片场景,八个主流 VLM 的攻击成功率能到七成。做视觉模型安全或部署多模态产品的人应该看看 S3TA 的攻击拆解思路和 SceneSafetyBench 的测试设计。
一篇新论文指出视觉语言模型存在场景状态安全漏洞:模型通过解读图像中的事件状态即可提供有害帮助,无需用户明确提出有害目标。团队构建了 SceneSafetyBench 基准,包含 375 组图像-问题对,覆盖初始、中期、终极三类场景状态和五种视觉呈现形式。在八个 VLM 上测试,场景状态输入的有害协助攻击成功率(ASR)达 45.6%-72.3%。
论文进一步提出 S3TA 系统化攻击方法,将有害目标拆解为场景状态、独立查询多组图像-问题对再拼装回答。在 MM-SafetyBench 七个类别、八个开源与商用 VLM 上,S3TA 取得 71.2% 的模型平均 ASR,比最强基线高出 6.1 个百分点。层级信号分析显示场景状态输入的安全信号特征与 QR 相关图像最接近,后期层读出值偏低。
文本侧过滤器和视觉审核对这类攻击防护效果有限。作者认为安全机制需要覆盖模型通过场景推理推断出的协助行为,而不只是筛查输入内容本身。
Reading Scene States: Uncovering Safety Failures in Vision-Language Models
Vision-language models (VLMs) can provide harmful assistance by interpreting visual events, even without an explicit request for the underlying harmful objective. We identify this \textbf{scene-state safety vulnerability} and introduce \textbf{SceneSafetyBench}, comprising 375 image--question pairs spanning Initial, Medium, and Ultimate scene states and five visual presentations. Across eight VLMs, scene-state inputs elicit harmful assistance with attack success rates (ASRs) of 45.6--72.3\%. Comparative analysis across three VLMs shows that scene-state inputs most closely resemble QR-related images in their layer-wise safety-related signal profiles, while exhibiting lower average late-layer readouts and distinct signal dynamics during generation. To turn this vulnerability into a systematic attack, we propose \textbf{S3TA}, which translates harmful objectives into scene states, queries image--question pairs independently, and assembles the responses. On seven MM-SafetyBench categories across eight open-weight and commercial VLMs, S3TA achieves a model-average ASR of \textbf{71.2\%, exceeding the strongest baseline by 6.1 percentage points}. Text-side filters and visual moderation provide limited protection. These findings highlight the need to safeguard the assistance models infer through scene-grounded reasoning, beyond screening input content.
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include SceneSafetyBench, VLM, AI safety, multimodal.
Source attribution: curated from arXiv cs.AI (@Tongxi Wu, Yuyang Cheng, Jiacheng Xian, Menglin Yang, Kai Chen, Jian Zhang, Yang Gao).