Another 240papers do not have category tags yet and are not included in filtered results.
提出 STCA，对驾驶视频 VLM 构造可迁移的时空对抗扰动
提出 LiBRA，在潜空间双向优化以规避图像水印双侧检测
AbstractLiBRA 用双向潜空间优化规避双侧水印检测，并提高相对反转基线的保真。
用通用扰动对恶意软件分类器及漂移检测器做规避与后门投毒
提出 HPAA 攻击，用排版视觉线索绕过内容审核系统
这篇论文研究了人类视觉感知与大模型分词处理机制之间的差异所导致的内容审核漏洞。
提出 CITA 三阶段框架，生成可绕过毒性检测器的中文隐式毒性文本
提出 FRA-Attack，用频域对齐与梯度正则做可迁移定向视觉攻击
FRA-Attack 是面向闭源 MLLM 的迁移式定向图像攻击。作者要解决的是：空间域局部对齐重复全局低频，代理梯度的高频成分又把更新拉向代理自身。
提出 Vaporizer，用词法变换、翻译和释义破解 LLM 输出水印
Proposes the universal curved mesh patch UCGP, causing cross-task semantic bias in infrared vision-language models.
UCGP is a universal physical adversarial patch for infrared vision-language models. The question is: can a patch that can be attached and reused across samples cause stable semantic deviations in the infrared visual evidence shared by classification, description, and VQA. The patch is constrained to a low-frequency curved mesh.
Extract system instructions with encoding and structured requests, and harden them by rewriting with CoT
This work evaluates and hardens the confidentiality of LLM system instructions under encoded extraction. The authors are concerned with: when direct extraction has been rejected, whether the model still surrenders protected content when the same request is rewritten as a formatting or encoding task.
Propose WebWeaver, which compromises only a single agent and recovers the multi-agent communication topology from conversation context.
WebWeaver is an inference attack against the confidentiality of LLM-MAS communication topology: it controls only one arbitrary agent and recovers the full graph from dialogue context, rather than querying agent IDs.
Propose GDA, which aligns image adversarial perturbations with text-grounded evidence and transfers attacks to vision-language models
GDA is a zero-query black-box encoder transfer attack for heterogeneous LVLMs.
提出 GradingAttack，用梯度后缀与角色扮演操纵教育评分智能体判定
提出 CAGE 攻击，在视觉 token 压缩下用图像扰动使预测偏离
CAGE 研究的是：LVLM 在推理时剪枝或合并视觉 token 之后，对抗稳健性该怎么估。
提出面向渗出的权重压缩，评估推理服务器权重窃取可行性
这篇工作把 LLM 权重渗出看成 APT：攻击者已有未加密权重，但必须把完整权重藏进推理 API 的合法输出。作者认为可压缩性是被忽略的因素，因为推理量化要求解压后仍能低开销前向，而偷权重的人可以接受事后微调。
提出字符级扰动与遗传算法去除 LLM 水印
这篇工作评估推理时 LLM 水印在黑盒、有限查询下的移除难度，并比较字符级、token 级和句子级编辑。
提出 AdvOF，生成对抗物体误导视觉语言导航智能体的场景感知
AdvOF 是针对 VLM 驱动 VLN 服务导航的对抗物体攻击。