Skip to content

Translation in progress Some entries still include text in the original language.

Oct 13 · Tuesday

All papers

Found 21 papers

Another 213papers do not have category tags yet and are not included in filtered results.

  1. DefensesarXiv 2610.09469

    Secure-CUA: Constraining the Impact of Untrusted Content in Computer-Use Agents

    提出 Secure-CUA,用动作事务约束计算机使用智能体的不可信影响

    Published
    Setting
    web agent
    Evaluated models
    Claude Opus 5, Gemini 3.8 Flash, GPT-5.6-Sol
    AbstractSecure-CUA 用每步事务约束不可信影响,良性 TSR 接近 Vanilla-CUA。

    Secure-CUA 要限制图形界面里不可信内容对 CUA 决策和 GUI 执行的影响,同时保留合法任务所需的不可信信息。

    Paper analysisRead analysis
  2. AttacksarXiv 2610.09240

    WebMirage hijacks visual web agents' browser operations with adversarial images

    提出 WebMirage,用局部对抗图像劫持视觉网页 Agent 的浏览器操作

    Published
    Setting
    web agent
    Evaluated models
    LLaVA-v1.5-13B, LLaVA-v1.6-34B, MiniCPM-o-8B and others6
    Abstract论文提出了针对网络智能体的端到端视觉红队框架 WebMirage,揭示了从视觉定位到浏览器执行的安全脆弱性。

    WebMirage 是一个面向视觉定位网络智能体的流水线感知端到端红队攻击框架。

    Paper analysisRead analysis
  3. Paper finds multimodal models' refusal failure rate rises by up to 68.7% after tool use is enabled

    评测 ReAct 工具调用对多模态模型拒答有害请求的削弱

    Published
    Setting
    AI Agent
    Evaluated models
    Gemini-2.5-Pro, Gemini-3.1-Pro-Preview, Claude Opus 4.6 and others13
    Abstract论文揭示了工具调用范式削弱 MLLM 拒答能力的现象,分析了稀释与偏移机制并验证了重注缓解方案。

    论文系统评估了智能体工具调用范式对多模态大语言模型拒答能力的影响,指出了工具增强视觉推理背后的安全隐患。核心探讨的问题在于:当 MLLM 从单次推理转向通过 ReAct 循环调用外部工具(打标、缩放、OCR、代码执行)时,工具使用机制本身是否会降低模型拒答有害请求的概率。

    Paper analysisRead analysis
  4. DrivingBench 发布:让视觉语言模型驾驶丰田 Corolla 绕桩

    提出 DrivingBench,评测通用视觉语言模型能否闭环驾驶真车

    Published
    Evaluated models
    GPT-6 Astra, Claude Fable 5.1, Grok 4.6 and others4
    Abstract仅 Astra 第二次跑完低速锥桶路线。

    DrivingBench 评测通用 VLM 能否自己闭环开真车,推理延迟也算进任务。

    Paper analysisRead analysis
  5. DIBench:面向移动 GUI 智能体多候选选择决策完整性的安全基准

    用 DIBench 评测移动 GUI 智能体被 UI 注入后的决策完整性

    Published
    Evaluated models
    Qwen2.5-VL(32B), Qwen3-VL(32B), GUI-Owl(32B) and others7

    Abstract论文提出移动智能体决策完整性基准 DIBench,发现欺诈 UI 注入诱导目标选择并推高完成率,通用防御收益不稳定。

    Paper analysisRead analysis
  6. MMSkillRisk:多模态技能图像中的隐藏指令可诱导 Agent 执行未授权操作

    提出 MMSkillRisk 基准,评测多模态技能图像中的隐藏指令诱导 Agent 越权操作

    Published
    Evaluated models
    GPT-5.6-sol, Kimi-K2.6, Kimi-K3 and others6

    Abstract论文提出 MMSkillRisk 基准与 NCVA 攻击,揭示多模态技能中教学图像潜藏指令可引发未授权操作并与良性任务共存。

    Paper analysisRead analysis
  7. MMPIBench:多模态提示注入对六种 Agent 框架的跨框架评测

    用 MMPIBench 评测多模态提示注入对 Agent 框架的影响

    Published
    Setting
    AI Agent
    Evaluated models
    Claude Opus 4.8, GPT-5.4, Gemini 3.1 Pro and others6

    Abstract论文提出 MMPIBench 评测多模态提示注入对智能体的影响,发现模型主导行为、尝试远超完成,音频通道缺乏防护。

    Paper analysisRead analysis
  8. AttacksarXiv 2609.04533

    Repeat-After-Me:针对黑盒 VLM 的自适应视觉提示注入攻击

    提出 Repeat-After-Me 自适应视觉提示注入,诱导 VLM 智能体执行指定任务

    Published
    Setting
    AI Agent
    Evaluated models
    Claude-Opus-4.7, GPT-5.5, Gemini-3.1-Pro and others6

    Abstract论文提出了针对黑盒 VLM 的前缀诱导自适应视觉提示注入,在多款模型上诱发了工具调用与隐私泄露。

    Paper analysisRead analysis
  9. 研究用因果审计揭示多模态模型视觉工具调用的假象

    因果审计视觉工具调用,分离观察贡献与动作诱发捷径

    Published
    Setting
    AI Agent
    Evaluated models
    DeepEyes, Pixel Reasoner, Mini-o3 and others6
    Abstract聚合增益集中在 Calibrated 少数轨迹,作者称之为视觉工具使用的错觉。

    对六个 thinking-with-images 模型做因果审计,检查 crop-and-zoom 返回的观察是否中介最终答案。

    Paper analysisRead analysis
  10. AttacksarXiv 2603.16960

    Adversarial attacks against modern vision-language models

    Conduct adversarial attacks on product images for open-source visual shopping agents and induce wrong purchases

    Published
    Setting
    web agent
    Evaluated models
    LLaVA-v1.5-7B, Qwen2.5-VL-7B
    AbstractThe authors state that LLaVA is easily induced by three attacks to buy the wrong item, while Qwen's correct purchase rate remains close to its clean condition.

    The authors compare the robustness of two open-source VLM shopping agents against image adversarial perturbations in a self-built e-commerce environment.

    Paper analysisRead analysis
  11. DefensesarXiv 2603.14707

    Visual Confusion Agents: Exploiting and Defending Against Perception Failures in Computer-Use Agents

    Proposes out-of-process dual-channel comparison classification to intercept erroneous clicks by computer-use agents

    Published
    Evaluated models
    multi-modal-embed-small
    AbstractPerception errors can be exploited by swapping screen pixels.

    The authors frame CUA's perception failure as a security vulnerability, visual confused deputy: the agent authorizes click(x,y) based on a misread screen, and the actual landing point differs from the element it believes it is clicking.

    Paper analysisRead analysis
  12. DefensesarXiv 2602.10915

    Blind Gods and Broken Screens: Building a Secure, Intent-Centric Mobile Agent Operating System

    Proposes Aura, which uses a trusted kernel to mediate the identity, perception, and action control of mobile agents

    Published
    Evaluated models
    Gemini-3-Flash
    AbstractAura mediates structured interactions via the Agent Kernel.

    Aura uses kernel-mediated structured agent interaction, replacing screen-interface mobile assistants..

    Paper analysisRead analysis
  13. WAInjectBench 系统评测网页 Agent 的提示注入检测方法

    提出 WAInjectBench 评测网页 Agent 的提示注入检测

    Published
    Setting
    web agent
    Evaluated models
    GPT-4o, Mistral-7B-Instruct-v0.1, Llama-Prompt-Guard-2-86M and others8
    AbstractWAInjectBench 比较网页智能体的提示注入检测:显式线索可被部分检出,隐蔽攻击大多漏检。

    WAInjectBench 是网页智能体提示注入检测的基准。作者要看现有检测器能否发现这类攻击:攻击者改网页,使智能体执行目标动作,而不是用户任务。作者称这是首个此类综合基准。

    Paper analysisRead analysis
  14. AttacksarXiv 2505.23266

    AdvOF:用对抗物体融合攻击 VLM 驱动的视觉语言导航智能体

    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

    Published
    Attacker
    White-box
    Evaluated models
    Lseg, Clip, ViT-L/16 and others5
    AbstractAdvOF 以多视角融合生成对抗物体。

    AdvOF 是针对 VLM 驱动 VLN 服务导航的对抗物体攻击。

    Paper analysisRead analysis
  15. AttacksarXiv 2505.11717

    WebInject:通过网页像素扰动对多模态网页 Agent 实施提示注入

    提出 WebInject,用网页像素扰动诱导网页 Agent 执行指定动作

    Published
    Setting
    web agent
    Evaluated models
    UI-TARS-7B-SFT, Phi-4-multimodal-instruct, Llama-3.2-11B-Vision-Instruct and others5
    AbstractWebInject 经源码扰动原始像素。

    WebInject 是对 MLLM 网页智能体的环境提示注入:修改网页源码,使原始像素上的扰动经过显示器映射后,诱导智能体执行攻击者指定动作。

    Paper analysisRead analysis