Skip to content

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

Oct 12 · Monday

All papers

Found 8 papers
Filter3

Attack type

Jailbreaks53Prompt injection22Poisoning and backdoors11Extraction and theft8Malicious use13

Defense type

Model hardening3

Impact

Data and privacy leaks6

System type

Models and APIs6

Attack surface / entry point

User input7Training data1

Layer

Training and data layer3
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 232papers have no tags yet and are not included in filtered results.
Attack type：Extraction and theftAttack surface / entry point：User inputAttack surface / entry point：Training dataClear all

Another 232papers do not have category tags yet and are not included in filtered results.

  1. Evaluation and benchmarksarXiv 2610.03434

    Persona Guardrail：面向智能体系统的生产级防御框架

    提出 Persona Guardrail，强制客服智能体守住功能边界

    arXiv
    2610.03434
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Attacker
    Black-box
    Evaluated models
    Qwen3-30B-A3B, Qwen3-VL-30B, Qwen3-8B and others5
    AbstractPersona Guardrail 用功能边界做同步输入输出校验，并在 PAGE 与生产延迟预算下给出评测。

    Persona Guardrail 是客服向智能体的同步运行时防御：先写明功能边界，再决定放行或拦截，而不是只检测某一攻击类。

    Paper analysisRead analysis
  3. Evaluation and benchmarksarXiv 2609.40286

    研究者提出 COVER：在语言预算下选择源语言实现跨语言遗忘

    提出 COVER，在语言预算下选择源语言实现跨语言知识遗忘

    arXiv
    2609.40286
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Tiny Aya Global, Qwen3-4B-Instruct-2507, Llama-3.2-3B-Instruct
    AbstractCOVER 在语言预算下选源语言，以降低未监督语言上相对 oracle 的残余访问。

    语言预算下的多语言遗忘：遗忘监督只放在 K 种源语言上，目标是降低其余语言对同一事实的残余访问，而不是把每种评测语言都译成监督数据。

    Paper analysisRead analysis
  5. Evaluation and benchmarksarXiv 2609.33481

    IDUnlearn-Bench：面向视觉语言模型的个体级遗忘基准

    提出 IDUnlearn-Bench，评测 VLM 个体级遗忘后的身份残留

    arXiv
    2609.33481
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-VL-2B, Qwen3-VL-4B, Qwen3-VL-8B and others6
    AbstractIDUnlearn-Bench 用四族探针显示：压低直接属性访问后，身份仍常能被反向查出、绑定或重构。

    IDUnlearn-Bench 评测 VLM 是否忘掉一个可跨模态识别的人，而不只是答不出某条监督事实。

    Paper analysisRead analysis
  7. Evaluation and benchmarksarXiv 2608.20554

    aiXamine：跨维度评测 LLM 安全、安全（security）与隐私权衡

    提出 aiXamine 黑盒基准，统一评测 LLM 安全、稳健与隐私权衡

    arXiv
    2608.20554
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    openai/gpt-5-mini, openai/gpt-4o, anthropic/claude-haiku-4.5 and others15
    AbstractaiXamine 用九服务黑盒协议比较百余个模型，作者称单轴信任度进展可能削弱其他轴。

    aiXamine 是一套黑盒评测协议，用来在同一条件下比较 LLM 的 safety、security 与 privacy，而不是再产出一个能力分。

    Paper analysisRead analysis
  8. Evaluation and benchmarksarXiv 2606.18673

    Understanding and mitigating prompt leakage attacks in real-world LLM applications

    Propose soft prompt defense AREA, re-anchoring attention to mitigate system prompt leakage

    arXiv
    2606.18673
    Published
    Layer
    Application layer
    Setting
    LLM applications
    Attacker
    Black-box
    Evaluated models
    Llama-2-7B-chat-hf, Llama-3.1-8B-Instruct, Mistral-7B-Instruct-v0.3 and others8
    AbstractThe authors measure commercial prompt leakage, propose attention drift, and use AREA to strike a trade-off between usability and leakage resistance.

    The authors study system prompt leakage in real LLM applications and propose AREA, a mitigation method that does not modify weights.

    Paper analysisRead analysis
You have reached the end