Skip to content

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

Oct 12 · Monday

All papers

Found 9 papers
Filter3

Attack type

Extraction and theft8

Impact

Harmful content141Data and privacy leaks9Output manipulation22Availability and reliability5Oversight failure4

System type

Models and APIs9

Attacker access

Black-box4

Attack surface / entry point

User input9

Layer

Prompt layer9Model layer18Application layer23Training and data layer9
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：Data and privacy leaksAttack surface / entry point：User inputLayer：Prompt layerClear all

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

  1. AttacksarXiv 2610.06093

    研究评测训练数据提取风险的跨语言迁移

    用翻译前缀做跨语言训练数据提取，恢复英语训练文本中的 PII

    arXiv
    2610.06093
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    GPT-Neo 1.3B, GPT-Neo 2.7B, GPT-J 6B and others7
    Abstract跨语言 TDE 能恢复英语记忆的 PII，转移随多语训练和措辞保真变化。

    这篇工作检验训练数据抽取是否跨语言：英语段落里记住的 PII，能否被其他语言的前缀逐字抽出。

    Paper analysisRead analysis
  3. Evaluation and benchmarksarXiv 2609.08256

    ACEA：面向红蓝队对抗的 LLM 协同演化测试平台

    提出红蓝对抗评测平台 ACEA，分开计量攻击强度与防御效果

    arXiv
    2609.08256
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    openai/gpt-4o, google/gemini-2.5-pro, google/gemma-3-27b-it and others5
    AbstractACEA 用 ASAP 让红蓝对同一目标对打。

    ACEA 是让可插拔红队和蓝队共享一个目标 LLM 的对打场地，把攻击强度和防御效果分开记分。作者要补三处空缺：双方只能对着固定对手分开练。LLM judge 分不清真实泄露和幻觉，拦截后又看不到攻击有多强。

    Paper analysisRead analysis
  4. AttacksarXiv 2608.09867

    Researchers steal proprietary LLM inference chains via cross-model compatibility of encrypted inference blocks

    利用加密思维块的跨模型重放，诱导弱模型逐字转写专有推理链

    arXiv
    2608.09867
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Claude Opus 4.8, Claude Sonnet 4.6, Claude Haiku 4.5 and others13

    Abstract论文揭示了客户端加密思维块跨模型互换漏洞，提出了低成本提取方法并给出防御方案。

    Paper analysisRead analysis
  5. AttacksarXiv 2604.01039

    评估与加固 LLM 系统指令以抵御编码攻击提取

    用编码与结构化请求抽取系统指令，并用 CoT 改写加固

    arXiv
    2604.01039
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    GPT-4.1-mini, GPT-3.5-turbo, Gemini-2.5-flash and others7
    Abstract编码式请求可在直接拒答后引出系统指令。

    这项工作评测并加固 LLM 系统指令在编码式抽取下的保密性。作者关心的是：直接抽取已被拒绝时，把同一请求改写成格式化或编码任务，模型是否仍交出受保护内容。

    Paper analysisRead analysis
  6. DefensesarXiv 2602.11528

    Stop Tracking Me! Proactive Defenses Against LLM Attribute Inference Attacks

    提出用户侧防御 TRACE-RPS，分享前匿名化文本并追加后缀以抑制属性推断

    arXiv
    2602.11528
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Llama2-7B-Chat, Llama2-13B-Chat, Llama3.1-8B-Instruct and others10
    Abstract用户侧 TRACE 加 RPS/MPS，在分享前降低 LLM 对敏感属性的推断准确率。

    这是一项用户侧防御：在文本公开前降低 LLM 对敏感属性的推断，而不是改攻击提示或削弱模型推理。

    Paper analysisRead analysis
  8. AttacksarXiv 2512.16059

    ContextLeak：审计私有上下文学习方法的隐私泄露

    提出 ContextLeak，用金丝雀成员推断审计私有 ICL 泄漏

    arXiv
    2512.16059
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Llama3.3-70b, Qwen2.5-72b, GPT-4.1 and others6
    AbstractContextLeak 审计私有 ICL 的 worst-case 泄漏。

    ContextLeak 是一套用成员推断审计私有 ICL 的框架，目标是给 worst-case 泄漏一个经验下界，而不是只引用理论 ε。

    Paper analysisRead analysis
You have reached the end