Skip to content

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

Oct 12 · Monday

All papers

Found 8 papers
Filter4

Attack type

Jailbreaks93Prompt injection54Poisoning and backdoors23Extraction and theft8Detection evasion33Malicious use7

Impact

Data and privacy leaks6

System type

Models and APIs4

Attacker access

Black-box3

Attack surface / entry point

User input5

Layer

Application layer4Training and data layer3

System component

MCP and skills3RAG7Guardrails and judges1Reasoning chains5Fine-tuning and open weights4Vision4Audio3Image generation5Embeddings5Multimodal12
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 233papers have no tags yet and are not included in filtered results.
Attack type：Extraction and theftSystem component：Guardrails and judgesSystem component：VisionSystem component：AudioClear all

Another 233papers do not have category tags yet and are not included in filtered results.

  2. DefensesarXiv 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
  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
  6. DefensesarXiv 2609.15671

    QPriv-VL：面向视觉语言模型的问题引导 token 裁剪隐私防御

    提出 QPriv-VL，用问题引导剪枝降低分布式视觉问答的隐私泄露

    arXiv
    2609.15671
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    LLaVA-1.5-7B, Llama-2
    AbstractQPriv-VL 在切层前按问题相关度与 DINOv2 敏感度剪视觉 token，作者称能在约四成 token 下兼顾准确率与四类攻击。

    QPriv-VL 把问题引导的视觉 token 剪枝做成分布式 VQA 的客户端隐私防御，而不是只做集中式推理加速。

    Paper analysisRead analysis
  7. AttacksarXiv 2609.01723

    研究者提出针对微调 TTS 模型的黑盒成员推断攻击

    提出针对微调 TTS 的黑盒成员推断攻击

    arXiv
    2609.01723
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    CosyVoice2, F5-TTS, XTTS-v2
    Abstract黑盒 TTS 成员推断以 recitation 加分层语音表示为分数，说话人级强于记录级，DP-SGD 可压到随机附近。

    作者给出一个针对微调文本转语音模型的黑盒成员推断框架，分别审计说话人是否进入微调集，以及某一条文本–音频记录是否进入微调集。微调 TTS 以合成文本和参考语音双条件生成，既有生成模型 MIA 的单条件查询和图像/文本上的逐点比较不能直接套用。作者用 scorable extent 与 memorization elicitation 比较五种查询，认为只有 recitation（用目标转写和干净目标语音要求模型复现）同时满足两条，并把它作为主查询。

    Paper analysisRead analysis
  8. AttacksarXiv 2602.07379

    Aegis: Governance, Integrity, and Security for AI Voice Agents

    提出 Aegis 红队框架，多轮攻击高风险语音 Agent

    arXiv
    2602.07379
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    GPT-4o, GPT-4o-mini, Gemini-1.5 Pro and others7
    AbstractAegis red-teams three types of voice agents: query-based access eliminates authentication and leakage successes, but behavioral attacks remain.

    Aegis uses multi-turn red-teaming evaluation on ALLM voice agents already integrated into business systems. The authors call this the first systematic red-teaming framework in this setting.

    Paper analysisRead analysis
You have reached the end