Skip to content

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

Oct 13 · Tuesday

All papers

Found 13 papers
Filter6

Attack type

Jailbreaks6Prompt injection4Extraction and theft3

Attack technique

Reasoning-chain manipulation4

Defense type

Inference-time intervention3

Impact

Harmful content6Data and privacy leaks4

System type

Models and APIs7AI Agent5

Attack surface / entry point

User input10Web pages and interfaces3Documents and messages3Tool responses4Weights and gradients1

Layer

Prompt layer14Model layer7Application layer6Training and data layer5

System component

MCP and skills23Memory31RAG19Guardrails and judges40Monitors28Reasoning chains13Permissions and sandboxing23Fine-tuning and open weights12Vision10Audio4Image generation7Embeddings4Multimodal20
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 215papers have no tags yet and are not included in filtered results.
Attack surface / entry point：User inputAttack surface / entry point：Documents and messagesAttack surface / entry point：Weights and gradientsLayer：Model layerLayer：Application layerSystem component：Reasoning chainsClear all

Another 215papers do not have category tags yet and are not included in filtered results.

  1. InterpretabilityarXiv 2610.12361

    Study: legal reasoning models citing legal provisions does not equal relying on legal provisions.

    Uses premise permutation and hidden-layer tracing to audit whether legal chain-of-thought truly relies on the cited legal provisions.

    arXiv
    2610.12361
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-8B, Qwen3-14B, Llama-3.1-8B and others7
    AbstractUsing counterfactual substitution of authorities and internal commitment tracing, the authors find that legal CoT rulings depend on authorities only to a limited extent and are vulnerable to prompt injection manipulation.

    This study conducts an empirical audit of the causal truthfulness of legal citations in chain-of-thought reasoning of large language models in the legal domain.

    Paper analysisRead analysis
  2. AttacksarXiv 2609.27542

    Researchers use control token injection to delete gpt-oss-20b's reasoning chain and bypass chain-of-thought monitoring

    提出控制标记注入，清空工具 Agent 的思维链以绕过推理监控

    arXiv
    2609.27542
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    openai/gpt-oss-20b, google/gemma-4-26B-A4B-it, deepseek-ai/DeepSeek-R1-Distill-Qwen-7B and others4

    Abstract论文阐明控制标记注入可抑制 CoT 使监控失效并促成拒绝绕过，且工具执行受解析器宽容度影响。

    Paper analysisRead analysis
  3. InterpretabilityarXiv 2609.23587

    大型推理模型的效率与安全困境：量化与剪枝如何影响越狱鲁棒性

    分析量化、剪枝与 KV 缓存压缩对推理模型越狱鲁棒性的影响

    arXiv
    2609.23587
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    QwQ-32B, DeepSeek-R1-Distill-Qwen-32B, DeepSeek-R1-Distill-Llama-8B
    Abstract效率压缩常使 LRM 越狱 HR 下降，作者将其归因于推理退化而非对齐增强。

    作者分析量化、剪枝和 KV 缓存压缩如何同时改变 LRM 的显存、推理和越狱脆弱性。。

    Paper analysisRead analysis
  5. OtherarXiv 2609.12413

    SoK 综述：智能体 AI 时代的越狱攻击、防御与实践考量

    系统梳理智能体执行链路中的越狱攻击与防御并比较推理时防御

    arXiv
    2609.12413
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Qwen3.5-4B, Llama-3.1-8B-Instruct
    AbstractSoK 将越狱重放到智能体流水线，并用两种模型检验回复层防御的边际与代价。

    Mia 等人把越狱安全从单轮回复扩展到智能体的规划、记忆、工具和智能体间通信，并问既有结论在更强原生对齐下是否还成立。

    Paper analysisRead analysis
  6. Risky behaviorarXiv 2609.08186

    论文揭示大推理模型的对齐崩溃并提出 Reasoning Trap 越狱范式

    揭示推理加深引发的对齐坍塌，用 Reasoning Trap 放大越狱

    arXiv
    2609.08186
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-8B, Qwen3-14B, Llama2-7B and others8
    Abstract更深推理提升题目准确率，同时提高扰动下的相对损失。

    作者研究扩展推理是否以对齐稳健性为代价，提出 ALR、RT 与 RRA。。

    Paper analysisRead analysis
  8. DefensesarXiv 2602.24210

    研究者提出 Staged Decoding 与 SFT 数据集，减少大推理模型推理轨迹中的隐私泄露

    提出 Staged Decoding 与指令跟随训练，减少推理轨迹隐私泄露

    arXiv
    2602.24210
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen 3 1.7B, Qwen 3 4B, Qwen 3 8B and others6
    Abstract用一般指令跟随训练加分段解码，可降低轨迹隐私泄露，但数学效用下降。

    本文把大推理模型的隐私问题做成轨迹可控性：隐私指令是指令，轨迹不跟随就会把敏感上下文写进可被提示注入导出的文本。

    Paper analysisRead analysis
  9. AttacksarXiv 2603.12277

    Prompt injection is role confusion

    提出 CoT Forgery，经用户消息或工具输出注入伪造推理

    arXiv
    2603.12277
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Attacker
    Black-box
    Evaluated models
    gpt-oss-20b, gpt-oss-120b, o4-mini and others11
    Abstract提示注入被作者解释为角色感知失败：听起来像某角色，就在表示里成为该角色。

    作者把提示注入解释成角色感知失败：标签划出的权限没有进入模型的内部几何。

    Paper analysisRead analysis
  10. AttacksarXiv 2602.18514

    招聘中的特洛伊木马：标准模型与推理模型间接提示注入的红队测试案例研究

    用简历中的不可见指令做间接提示注入，操纵招聘推荐

    arXiv
    2602.18514
    Published
    Layer
    Application layer
    Setting
    LLM applications
    Evaluated models
    Qwen 3 30B A3B Instruct 2507, Qwen 3 30B A3B Thinking 2507

    Abstract作者称推理在简单注入下会帮攻击圆谎，在不合逻辑的复杂注入下又会把欺骗漏出来。

    Paper analysisRead analysis
  11. DefensesarXiv 2602.11096

    Safety Recovery of Reasoning Models Only Needs a Few Early Steps of Guidance

    Proposes SafeThink, an inference-time defense that monitors the chain of thought and injects a redirect prefix when it falls below a threshold.

    arXiv
    2602.11096
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    R1-Onevision-7B, OpenVLThinker-7B, VLAA-Thinker-7B and others6
    Paper analysisRead analysis
  12. DefensesarXiv 2601.09281

    STaR: A sensitive trajectory regulation method for unlearning in large reasoning models

    提出推理时框架 STaR，抑制大推理模型推理轨迹中的敏感内容

    arXiv
    2601.09281
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    DeepSeekR1-Distill-Llama-8B
    AbstractSTaR 以推理时轨迹调控做 LRM unlearning，作者称在 R-TOFU 上遗忘与隐私指标更好。

    STaR在推理阶段约束大推理模型的整段 CoT，使指定 forget set 中的敏感内容不出现在中间步骤或最终答案里，且不更新参数。问题来自 LRM 会把敏感知识写进多步推理，而答案级遗忘在 ZeroThink、LessThink 等解码下仍可能泄漏。STaR 先判断查询是否相关并检索遗忘片段，再加安全前缀。

    Paper analysisRead analysis
You have reached the end