Skip to content

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

Oct 12 · Monday

Agent safety · Papers

Found 3 papers
Filter2

Attack type

Prompt injection12Poisoning and backdoors3

Impact

Unauthorized actions3

System type

AI Agent3

Attack surface / entry point

User input6Web pages and interfaces3Documents and messages3Tool responses5Tool and skill descriptions7Memory writes9Distribution channels7

Layer

Application layer3

System component

Permissions and sandboxing3
Select any value within a category; different categories must all match. Counts include the other filters you have selected.
Attack type：Poisoning and backdoorsAttack surface / entry point：Documents and messagesTopic: Agent safetyClear all
  1. DefensesarXiv 2610.01349

    PACE：面向工具调用 LLM Agent 的来源感知能力强制执行

    提出 PACE，在工具调用前做路径隔离与效果授权

    arXiv
    2610.01349
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    gpt-5.6-luna, DeepSeek-V4-Flash, Qwen-3.8-27B
    AbstractPACE 在工具边界做路径割与效果核验。

    PACE 是面向工具调用型 LLM 智能体的运行时调解：当准入证据无法区分会改变行为的规格时，不在出口站点上撤销调解，而在最后一次工具调用前同时检查表示出的影响路径和请求权限。

    Paper analysisRead analysis
  2. 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
  3. DefensesarXiv 2603.12644

    揭示自主 Agent 的安全威胁与防御架构：OpenClaw 案例研究

    梳理自主 Agent 威胁并提出全生命周期防御蓝图 FASA

    arXiv
    2603.12644
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Abstract以 OpenClaw 为例划分三层风险，并用 FASA 描述从输入到系统执行的防御。

    这是一份以 OpenClaw 为案例的威胁梳理和防御架构报告，不是带分数的攻击或防御评测。

    Paper analysisRead analysis
You have reached the end