Skip to content

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

Oct 12 · Monday

All papers

Found 7 papers
Filter5

Attack type

Poisoning and backdoors5Detection evasion3

Impact

Output manipulation7

System type

Models and APIs7

Attack surface / entry point

User input7Documents and messages2Memory writes0Images, audio and video11Training data5Weights and gradients5Distribution channels3Inference services and hardware3

Layer

Model layer7Application layer33Training and data layer24
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.
Impact：Output manipulationAttack surface / entry point：Documents and messagesAttack surface / entry point：Memory writesAttack surface / entry point：Training dataLayer：Model layerClear all

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

  1. DefensesarXiv 2610.00348

    NEEDLE：通过权重正交化移除 LLM 后门

    提出 NEEDLE，用权重正交化在已知触发器时去除大模型后门

    arXiv
    2610.00348
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Gemma-3-4B-IT, Qwen3-4B-Instruct-2507, Gemma-3-12B-IT

    AbstractNEEDLE 用闭式权重编辑去掉已知触发器的后门，并保住拒绝子空间。

    Paper analysisRead analysis
  2. DefensesarXiv 2609.00786

    MROP：针对 Deep JSCC 后门攻击的掩码区域优化净化方法

    提出 MROP，在发射端推理时纯化输入以抵御训练投毒后门

    arXiv
    2609.00786
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    DeepJSCC, SwinJSCC
    AbstractMROP 不重训练，在发射端纯化补丁后门。

    MROP 是 deep JSCC 无线图像传输上的推理时输入纯化，对象是矩形补丁后门。

    Paper analysisRead analysis
  5. Evaluation and benchmarksarXiv 2607.24177

    EXE-Bench：为真实可用性对 AI Windows 恶意软件检测器进行权衡排名

    提出 EXE-Bench 比较 Windows 恶意软件检测器的可用性权衡

    arXiv
    2607.24177
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    EmberGBDT, BBDnn, BBDnnFDRS and others15
    Abstract统一四指标后，手工特征的 EmberGBDT 综合居首，只看准确率会误导部署。

    EXE-Bench 用同一设置给 AI Windows 恶意软件检测器做四指标排序，而不是只比一次测试集准确率。。

    Paper analysisRead analysis
  6. DefensesarXiv 2603.03206

    理解并缓解大语言模型引导中的数据集污染

    用 Lee-Valiant 稳健均值缓解 steering 数据集污染

    arXiv
    2603.03206
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.2-3B-Instruct, Mistral-7b-Instruct-v0.3, OLMo-2-1124-7B-Instruct
    Abstract数据污染会改 steering，协调注入最强。

    作者研究 contrastive steering 的训练数据被污染后，推理时的行为控制还剩多少，以及稳健均值能否接住。

    Paper analysisRead analysis
  7. DefensesarXiv 2602.03085

    Triggers in the Haystack: Extracting and Reconstructing LLM Backdoor Triggers

    提出推理扫描器，在仅有模型文件时检测后门并重构触发词

    arXiv
    2602.03085
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    gemma-3-270m-it, DeepSeek-R1-Distill-Qwen-1.5B, Phi-4-mini-instruct and others8
    Abstract用记忆泄漏和内部信号重构触发词。

    这是一个只做推理的 sleeper agent 扫描器：在不知触发词和目标行为时，检测因果语言模型里的条件后门并重构触发词。。问题在于，后门只在触发词出现时偏离基线，而开放仓库和网页投毒使检测方通常没有触发词、目标或激活语料。

    Paper analysisRead analysis
You have reached the end