Skip to content

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

Oct 12 · Monday

All papers

Found 17 papers
Filter3

Attack type

Jailbreaks17Prompt injection47Poisoning and backdoors13Extraction and theft4Detection evasion3

Attack technique

Cross-modal payloads5

Defense type

Permissions and approval5

Impact

Harmful content10Unauthorized actions4

System type

Models and APIs11AI Agent5

Attacker access

Black-box6White-box3

Attack surface / entry point

User input11Images, audio and video11

Layer

Prompt layer9Application layer6

System component

Memory3Guardrails and judges55Monitors11Reasoning chains15Permissions and sandboxing5Fine-tuning and open weights4Vision35Audio12Image generation15Multimodal61
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：JailbreaksSystem component：Permissions and sandboxingSystem component：AudioClear all

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

  1. Analysis and theoryarXiv 2610.07005

    研究审计 Qwen2-Audio 音频越狱 AdvWave-P 的频率与解码层归因

    遮挡审计 AdvWave-P 音频越狱的频带必要性与解码层关联

    arXiv
    2610.07005
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2-Audio-7B-Instruct, LLaMA-Omni
    AbstractAdvWave-P 的八频带排序与能量相关。

    这是对加性音频越狱 AdvWave-P 的频率与解码器深度审计，受害模型主要是 Qwen2-Audio-7B-Instruct，不提出新攻击。

    Paper analysisRead analysis
  2. DefensesarXiv 2609.34518

    SEAD 论文提出工具型 Agent 的状态视角攻击与防御方法 DART 与 SAGE

    提出工具型 Agent 的自适应攻击 DART 与预执行防御 SAGE

    arXiv
    2609.34518
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    GPT-5.6 Luna, GPT-5.6 Terra, Gemini 3.8 Flash and others5

    Abstract论文提出状态控制框架 SEAD、攻击 DART 与防御 SAGE，实验显示执行反馈与环境探测在攻防交互中具有关键作用。

    Paper analysisRead analysis
  4. DefensesarXiv 2609.28693

    skilder：以角色化技能包为工具型 LLM 智能体做访问控制

    提出 Skilder，按角色渐进下发企业工具并确定性拦截越权调用

    arXiv
    2609.28693
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Claude Haiku 4.5, Qwen 3.5 122B, Gemma 4 31B and others6
    AbstractSkilder 用单 MCP 服务器按角色交付工具并硬拦截。

    Skilder 把企业工具型智能体的能力按角色打包，经一个 MCP 服务器做渐进发现，使工具范围由服务器而不是提示词来执行。

    Paper analysisRead analysis
  7. DefensesarXiv 2609.14003

    FlowSeal：用信息流控制阻断 LLM Agent 的隐私泄露

    提出 FLOWSEAL，用工具级信息流控制阻断智能体隐私泄露

    arXiv
    2609.14003
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    DeepSeek-V3.2, GPT-4.1-mini, Gemini-Flash-2.5

    Abstract作者分析大模型智能体隐私检查与攻击面重合问题，提出外部拦截防御 FLOWSEAL 并验证有效性。

    Paper analysisRead analysis
  10. Evaluation and benchmarksarXiv 2606.06037

    SpeechJBB：语码转换语音下大型音频语言模型的越狱基准

    提出 SpeechJBB，评测语码转换语音下音频语言模型的越狱稳健性

    arXiv
    2606.06037
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2.5-Omni-7B, Qwen3-Omni-30B-A3B-Instruct, Voxtral-Small-24B and others9
    AbstractSpeechJBB 用五语语码转换语音和伪词探针评测九个 LALM，作者将越狱归为对齐失败而非听不懂。

    SpeechJBB 用合成语音测量 LALM 在五语、语码转换和局部伪词混淆下的安全对齐，并另设口语理解对照。

    Paper analysisRead analysis
  15. AttacksarXiv 2511.10692

    StyleBreak：利用风格化音频越狱揭示大型音频语言模型的对齐漏洞

    提出 StyleBreak 风格化音频越狱，利用语音属性绕过音频语言模型对齐

    arXiv
    2511.10692
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Qwen2-Audio-7B-Instruct, MERaLiON-AudioLLM-Whisper-SEA-LION, Ultravox-v0.4.1-Llama-3.1-8B and others6
    Paper analysisRead analysis
  16. AttacksarXiv 2504.01094

    研究者提出 Multi-AudioJail

    提出 MULTI-AUDIOJAIL，以多语多口音和声学扰动越狱音频模型

    arXiv
    2504.01094
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Qwen2-Audio, DiVA-llama-3-v0-8b, MERaLiON-AudioLLM-Whisper-SEA-LION and others5

    Abstract黑盒多语多口音音频加声学扰动可抬高 LALM 的 JSR，文本示范防御并非在每个模型上都下降。

    Paper analysisRead analysis
You have reached the end