Skip to content

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

Oct 12 · Monday

All papers

Found 1 paper
Filter2

Defense type

Model hardening3Inference-time intervention1Detection and filtering10Instruction–data separation12Permissions and approval10Monitoring and auditing7Watermarking and provenance8

Attack surface / entry point

User input22Documents and messages1Images, audio and video6Training data3
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 240papers have no tags yet and are not included in filtered results.
Defense type：Inference-time interventionAttack surface / entry point：Documents and messagesClear all

Another 240papers do not have category tags yet and are not included in filtered results.

  1. DefensesarXiv 2602.20708

    ICON：基于推理时纠正的 Agent 间接提示注入防御

    提出推理时防御 ICON，探测注意力过聚焦并转向以纠正 Agent 间接提示注入

    arXiv
    2602.20708
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Qwen-3-8B, Llama-3.1-8B, Mistral-8B and others6
    AbstractICON 在推理时检测过聚焦并转向，以兼顾 IPI 安全与任务效用。

    ICON 在推理时检测 IPI 的注意力过聚焦并转向，避免以拒答中止智能体轨迹。

    Paper analysisRead analysis
You have reached the end