Skip to content

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

Oct 12 · Monday

All papers

Found 14 papers
Filter4

Attack type

Jailbreaks9Poisoning and backdoors3

Attack technique

Gradient and representation optimization48Automated search53Multi-turn escalation22Encoding and obfuscation44Role-play and persuasion24Conversation structure manipulation10Reasoning-chain manipulation11Cross-modal payloads47Defense-targeted15Dormant triggers4

Impact

Harmful content9Output manipulation3

System type

Models and APIs14

Attacker access

Black-box7

Attack surface / entry point

User input11

Layer

Prompt layer10Model layer4Application layer39Training and data layer19
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 247papers have no tags yet and are not included in filtered results.
Attack technique：Conversation structure manipulationAttack technique：Dormant triggersLayer：Prompt layerLayer：Model layerClear all

Another 247papers do not have category tags yet and are not included in filtered results.

  1. AttacksarXiv 2610.13928

    AI models can provably hide arbitrary capabilities

    Proposes digital lock compilation, encrypting executable networks into host weights to hide arbitrary capabilities

    arXiv
    2610.13928
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Gemma 3 270M
    AbstractThe authors use a digital lock to hide an executable capability and state that elicitation on the chemistry and parity tasks did not recover the capability.

    The authors construct a class of backdoors that encrypt an executable network into the weights, used to discuss the limits of white-box elicitation audits under supply chain attacks.

    Paper analysisRead analysis
  2. AttacksarXiv 2610.06049

    研究者提出仅修改 SAE 解码器的后门攻击，可在不改动语言模型的情况下植入行为

    提出只改 SAE 解码器的后门，在冻结语言模型上植入代码插入

    arXiv
    2610.06049
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Gemma 3 4B Instruct, Qwen3 1.7B, Llama 3.1 8B Instruct
    Abstract只改解码器的 SAE 可带上代码插入后门，且若干常规质量指标变化不大。

    只改 SAE 解码器、冻结编码器与语言模型，即可把代码插入行为放进推理时的辅助组件。作者把这视为 SAE 供应链问题：checkpoint 一旦替换某一层激活，就参与后续计算，而不只是解释工具。

    Paper analysisRead analysis
  3. AttacksarXiv 2609.36956

    研究者提出 Controlled Decoding Attacks，通过纯文本接口越狱黑盒 LLM

    提出 BLINDBIAS，通过纯文本续写与前缀控制越狱黑盒模型

    arXiv
    2609.36956
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    GLM-5, Gemini-3.5-Flash, Qwen3-32B and others4
    Abstract作者认为，隐藏数值概率并不足以消除支持重复采样和前缀续写的解码攻击面。

    BLINDBIAS 是一种针对 sample-only 续写接口的 black-box 解码时越狱框架，研究没有目标数值概率时，采样能否支持逐步控制。

    Paper analysisRead analysis
  6. 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
  9. Analysis and theoryarXiv 2606.04168

    论文提出随机插入攻击，利用自回归一致性绕过 LLM 安全对齐

    解释自回归一致性如何导致浅层安全对齐，并提出随机插入攻击

    arXiv
    2606.04168
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Abstract作者用自回归一致性串起浅层对齐、插入攻击和最坏延续对齐。

    作者把浅层安全对齐和一类续写攻击都归到自回归一致性上。

    Paper analysisRead analysis
  10. InterpretabilityarXiv 2603.08234

    续写与拒答之争：大语言模型中续写触发越狱的机制分析

    分析续写触发越狱的注意力头机制并提出 HCS

    arXiv
    2603.08234
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    LLaMA-2-7B-Chat, Qwen2.5-7B-Instruct
    Abstract作者称越狱来自两类头竞争，HCS 加蒸馏可把该竞争用于推理时干预。

    作者用注意力头机制分析 continuation-triggered jailbreak：同一有害指令下，把 continuation 后缀从指令后面挪到提示词边界外，模型更容易顺着后缀继续生成。

    Paper analysisRead analysis
  11. AttacksarXiv 2602.15927

    研究者提出面向多轮对话的视觉记忆注入攻击 VMI

    提出 VMI，用扰动图像在多轮对话中潜伏并仅在触发话题输出预定消息

    arXiv
    2602.15927
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2.5-VL-7B-Instruct, Qwen3-VL-8B-Instruct, LLaVA-OneVision-1.5-8B-Instruct and others5
    AbstractVMI 让扰动图像在多轮 LVLM 中潜伏，作者称触发话题出现后仍可输出预定消息。

    VMI 是面向多轮 LVLM 的定向图像攻击：第三方发布带小扰动的图像，良性用户把它当作普通输入。

    Paper analysisRead analysis
  13. AttacksarXiv 2510.10271

    MetaBreak：通过特殊 token 操纵越狱在线 LLM 服务

    提出 MetaBreak，用特殊 token 操纵越狱在线 LLM 服务并绕过审核

    arXiv
    2510.10271
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Llama-3.3-70B-Instruct, Qwen-2.5-72B-Instruct, Gemma-2-27B-Instruct and others10

    Abstract利用特殊 token 注入对抗平台模板包裹与审核，多场景验证了越狱有效性与互补性。

    Paper analysisRead analysis
You have reached the end