Skip to content

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

Oct 12 · Monday

All papers

Found 48 papers
Filter3

Attack type

Jailbreaks48

Attack technique

Gradient and representation optimization24Automated search46Multi-turn escalation22Encoding and obfuscation37Role-play and persuasion17Conversation structure manipulation9Reasoning-chain manipulation6Cross-modal payloads32Defense-targeted9Dormant triggers4

Defense type

Detection and filtering3

Impact

Harmful content48Unauthorized actions7Output manipulation27Availability and reliability4

System type

Models and APIs46

Attacker access

Black-box36White-box3

Attack surface / entry point

User input43Images, audio and video32

Layer

Prompt layer45

System component

Guardrails and judges11Reasoning chains3Vision24Audio5Image generation4Multimodal32
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.
Attack technique：Role-play and persuasionAttack technique：Cross-modal payloadsImpact：Harmful contentClear all

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

  3. AttacksarXiv 2609.07216

    DIVA：利用视觉锚点对视频生成模型实施多模态越狱

    提出 DIVA，将有害意图拆成视觉锚点与运动文本以越狱视频生成模型

    arXiv
    2609.07216
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Wan, CogVideoX, LTX-Video and others9
    AbstractDIVA 用视觉锚点抑制语义漂移做视频越狱，并给出多条件安全基准 TI2VSafetyBench。

    DIVA 是针对带参考图的视频生成模型的免训练 black-box 越狱：用一致性机制本身堵住语义漂移这条安全逃逸路径。。

    Paper analysisRead analysis
  4. AttacksarXiv 2609.02414

    Blueprint 框架用世界观模拟与 18 个影响因子提升多轮越狱成功率

    提出多轮越狱框架 BLUEPRINT，用世界观模拟与心理因素诱导有害输出

    arXiv
    2609.02414
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Qwen3-Next-80B, DeepSeek-V3.2, GLM-4.7 and others6
    AbstractBLUEPRINT 解耦多轮越狱策略与情境模拟，揭示任务具象化是摆脱拒绝的核心机制，为多轮安全防御提供新视角。

    BLUEPRINT 是一个将因子化社会影响力策略空间与跨轮世界观模拟解耦的多轮大语言模型安全评测框架。

    Paper analysisRead analysis
  9. AttacksarXiv 2607.23496

    Concept2Scenario：用 SAE 概念归因发现可迁移的越狱场景

    提出 Concept2Scenario，借助内部概念归因发现可迁移的越狱场景

    arXiv
    2607.23496
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    White-box, Black-box
    Evaluated models
    Qwen3.5-9B, Llama-3.1-8B, Ministral-3-8B and others6

    Abstract该研究从表征空间因果归因揭示场景削弱拒绝的机制，发现可复用脆弱场景先验，提升多模型黑盒越狱效能与探索速度。

    Paper analysisRead analysis
  10. AttacksarXiv 2607.02961

    研究者提出信息过载方法越狱大型视觉语言模型

    提出 INFER+ 信息过载方法，越狱大型视觉语言模型

    arXiv
    2607.02961
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box, Gray-box
    Evaluated models
    Qwen3-VL-8B, Qwen2-VL-7B, InternVL3.5-8B and others8
    Abstract用图文过载越狱 LVLM。

    这篇工作用信息过载越狱 LVLM，而不是靠 OOD 变换隐藏意图。

    Paper analysisRead analysis
  13. AttacksarXiv 2607.21619

    ASO：用GRPO优化视觉风格放大MLLM越狱攻击

    提出 ASO，用 GRPO 优化视觉风格以放大越狱攻击

    arXiv
    2607.21619
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Gray-box
    Evaluated models
    GPT-4.1-mini, Gemini-2.5-Flash, Qwen3-VL and others6

    Abstract论文揭示了 MLLM 的风格敏感性脆弱面，提出 ASO 框架利用强化学习优化风格触发器，有效放大了现有视觉越狱效果。

    Paper analysisRead analysis
  14. AttacksarXiv 2606.01837

    论文提出 DSR 跨模态越狱框架：用良性输入诱导 MLLM 有害输出

    提出 DSR 跨模态越狱，用无害图文诱导 MLLM 生成有害图像

    arXiv
    2606.01837
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    gemini-2.5-flash-image, qwen-image-2.0, Tongyi Wanx 2.5
    AbstractDSR 用无害图文诱导商业 MLLM 输出有害图像，作者称组合能力会反过来扩大攻击面。

    DSR 是一种 black-box 跨模态越狱：输入侧不放显式有害内容，而让商业 MLLM 在推理中把无害图文重组成有害图像。

    Paper analysisRead analysis
  15. Evaluation and benchmarksarXiv 2606.02630

    MultiTurnPSB

    构建 MultiTurnPSB，评测四轮医疗越狱与分类器输入防御

    arXiv
    2606.02630
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    GPT-4.1-mini, Claude Sonnet 4.5
    Abstract多轮攻击加剧医疗模型风险，分类器干预有效但受高误报率制约。

    MultiTurnPSB 针对患者端医疗大模型在持续交互中的安全风险，构建了包含三种攻击模态的四轮对抗评测基准。

    Paper analysisRead analysis
  18. AttacksarXiv 2605.07399

    研究者提出 GPO-V，用全局概率优化越狱扩散视觉语言模型

    提出 GPO-V，用全局概率优化的视觉扰动越狱扩散视觉语言模型

    arXiv
    2605.07399
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    White-box, Black-box
    Evaluated models
    LLaDA-V, LaViDA
    Abstract作者用视觉扰动改 dVLM 的全局生成概率，以绕过两类拒绝。

    GPO-V 把越狱从“开头必须顺从”改成“整段去噪概率朝肯定方向走”。。

    Paper analysisRead analysis
  20. AttacksarXiv 2605.00583

    Jailbreaking vision-language models through the visual modality

    提出四类视觉越狱，把有害意图编码进图像以绕过安全拒答

    arXiv
    2605.00583
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    GPT-5.2, Claude-Haiku-4.5, Gemini-3-Flash and others8
    Abstract四类视觉越狱在六个 VLM 上检验文本对齐能否挡住视觉意图，模型与攻击之间差异大。

    四类视觉越狱用来检验前沿 VLM 的文本安全训练能否挡住由图像传达的有害意图。。

    Paper analysisRead analysis
Load more