Skip to content

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

Oct 12 · Monday

All papers

Found 8 papers
Filter2

Attack type

Prompt injection3

Model risks

Dangerous agent actions4

Defense type

Permissions and approval3

Impact

Unauthorized actions8Data and privacy leaks3

System type

AI Agent8

Layer

Application layer8

System component

MCP and skills7Memory6Guardrails and judges7Monitors5Permissions and sandboxing8Vision6Multimodal7
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 265papers have no tags yet and are not included in filtered results.
Impact：Unauthorized actionsSystem component：Permissions and sandboxingClear all

Another 265papers do not have category tags yet and are not included in filtered results.

  2. Evaluation and benchmarksarXiv 2610.07359

    论文实测 Agent 工具调用门控栈的失败相关性

    提出 nmult 并评测 Agent 动作门禁堆叠的错误相关性

    arXiv
    2610.07359
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    AgentTrust runtime (174 rules), GPT-5.5 (gpt55), claude-haiku-4-5-20251001 and others7
    Abstract论文实测发现大模型裁判间错误显著相关，作者主张按机制选型并在整栈层面度量防护增益。

    论文是一项针对智能体动作拦截防御堆叠故障关联性与实际组合价值的实证测量研究。

    Paper analysisRead analysis
  4. Evaluation and benchmarksarXiv 2610.02664

    论文提出 GHOST：长程智能体在良性对话中遗忘早期安全约束

    提出 STAR-Guard 双层防御，抑制长程 Agent 遗忘跨轮安全约束

    arXiv
    2610.02664
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    GPT-5.5, DeepSeek-V4-Pro, GLM-5.2 and others7

    Abstract揭示 GHOST 跨轮约束遗忘风险，构建 SCARBench 基准并通过 STAR-Guard 双层防御消除违规。

    Paper analysisRead analysis
  5. Evaluation and benchmarksarXiv 2609.26900

    Ajar：在 AgentDojo 上测量 Agent 防御中未关闭的权限

    提出 Ajar，测量 Agent 防御对任务不需要的工具调用的放行

    arXiv
    2609.26900
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Sonnet-5, Haiku-4.5
    AbstractAjar 用宿主基准的真值给防御的开放权限打分，OPL 独立于攻击成功和效用。

    Ajar 量的是智能体防御留下的开放权限：任务不需要、但防御仍会允许的工具调用。

    Paper analysisRead analysis
  7. Evaluation and benchmarksarXiv 2608.09476

    ActBench：面向协作智能体行为安全的自演化基准

    构建自演化基准 ActBench，评测协作智能体的行为安全

    arXiv
    2608.09476
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Claude-Opus-4.8, Claude-Sonnet-4.6, GPT-5.5 and others15

    AbstractActBench 通过自演化搜索与双证据重构评测轨迹行为安全，揭示基础模型决策差异主导了协同智能体的操作风险。

    Paper analysisRead analysis
  8. Evaluation and benchmarksarXiv 2608.07556

    MasDrift：跨多智能体架构的授权保持基准

    提出 MasDrift，评测多智能体在良性任务中的授权保持

    arXiv
    2608.07556
    Published
    Layer
    Application layer
    Setting
    Multi-agent systems
    Evaluated models
    DeepSeek V4 Flash, Qwen3.7 Plus, GPT-5.4 Nano and others6
    Abstract授权保持可被测量：层级用未授权动作换完成率，重锚定比沿链携带更稳。

    MasDrift 把多智能体设计中的授权保持做成可测量性质，对象是没有攻击者的普通委托。

    Paper analysisRead analysis
You have reached the end