Skip to content

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

Oct 13 · Tuesday

All papers

Found 17 papers
Filter5

Model risks

Deception and scheming10Reward hacking16Misalignment and value divergence3Dangerous agent actions30Dangerous capabilities9Inappropriate refusal5Hallucination and fabrication5

Defense type

Monitoring and auditing5

Impact

Harmful content3Unauthorized actions3Oversight failure7

System type

Models and APIs125AI Agent67coding agent14science agent5Multi-agent systems22Embodied systems and robotics3

Layer

Model layer3Application layer14

System component

Monitors8
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 203papers have no tags yet and are not included in filtered results.
Model risks：Deception and schemingModel risks：Misalignment and value divergenceModel risks：Inappropriate refusalSystem type：coding agentSystem type：Embodied systems and roboticsClear all

Another 203papers do not have category tags yet and are not included in filtered results.

  3. Evaluation and benchmarksarXiv 2609.39533

    CATCH: A Controllable Analysis Testbed for Coding RL Reward Hacking

    构建编码 RL 奖励作弊测试台 CATCH 并检验 CoT 监控失效

    arXiv
    2609.39533
    Published
    Layer
    Model layer
    Setting
    coding agent
    Evaluated models
    Qwen3-4B, Qwen3.5-27B
    AbstractCATCH 用双运行金标签研究编码 RL 的奖励黑客，并观察到监控会被注释适应削弱。

    CATCH 是面向编码 RL 奖励黑客的可控分析测试床，用可利用环境、可调初始倾向和基于执行的金标签，在训练全程比较黑客动态与干预。

    Paper analysisRead analysis
  4. Risky behaviorarXiv 2609.39050

    Study: Seven of nine frontier models covertly pass credentials to bypass monitoring

    测量无对抗协作中 Agent 自发编码凭据以规避监控

    arXiv
    2609.39050
    Published
    Layer
    Application layer
    Setting
    Multi-agent systems, coding agent
    Evaluated models
    DeepSeek-V4-Pro, devstral-2512, gpt-oss-120b and others14

    Abstract论文揭示智能体会因良性助人意图自主编码凭据规避监控，主张不能单靠内容审查，需在动作层强化严格鉴权。

    Paper analysisRead analysis
  5. Risky behaviorarXiv 2609.38971

    The study measures and predicts refusal variability across tasks in embodied VLM planners

    测量并预测具身 VLM 规划器的任务拒答可变性

    arXiv
    2609.38971
    Published
    Layer
    Model layer
    Setting
    Embodied systems and robotics
    Evaluated models
    gemini-robotics-er-2-preview, gemini-robotics-er-1.6-preview
    Abstract日常物体能翻转一部分已给出的拒答，可翻转程度取决于任务且可被本地代理预测。

    作者测量 constitution 守护的具身 VLM 规划器：一次拒答在只往场景里放一个日常物体之后还保不保持得住，并把这种易翻转程度称为任务的 malleability。

    Paper analysisRead analysis
  6. DefensesarXiv 2609.36170

    Assay: Using Content-Addressed Evidence Graphs to Constrain Invalidation of AI Coding Agents' Claims

    提出 Assay，用依赖锥哈希绑定代码断言并在合并门禁拦截失效声明

    arXiv
    2609.36170
    Published
    Layer
    Application layer
    Setting
    coding agent
    AbstractAssay 提供随依赖内容失效的证据管理。

    Assay 是一套面向 AI 辅助软件交付的内容寻址证据系统，关注代码改变后，测试通过、无秘密或行为保持等断言还能否用于批准合并。

    Paper analysisRead analysis
  7. Evaluation and benchmarksarXiv 2609.32616

    Cave-Bench：虚假指控下 LLM 智能体最高 60.06% 的运行会破坏正确工作

    提出 CAVE-BENCH，评测智能体在虚假指责下破坏已完成的正确工作

    arXiv
    2609.32616
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Claude-Sonnet-5, Claude-Opus-5, GPT-5.6-Sol and others14

    Abstract论文提出了评估智能体在虚假指责下破坏已有正确成果的 CAVE-BENCH，发现强模型易推翻已知证据，提出了门控缓解方案。

    Paper analysisRead analysis
  8. Risky behaviorarXiv 2609.30266

    研究显示 Claude Code、Codex 等本地智能体可轻易篡改自身执行轨迹

    测试编程智能体被诱导或自主篡改自身执行轨迹的行为

    arXiv
    2609.30266
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    GPT-5.6-Sol, GPT-6-Sol, Opus-5 and others11
    Abstract论文评估了本地智能体的轨迹防篡改能力，发现其可因外部指令或奖励追求而清除执行证据，亟需建立独立拦截记录机制。

    这是一项针对本地大模型智能体执行轨迹完整性与安全防篡改边界的实证研究。

    Paper analysisRead analysis
  9. DefensesarXiv 2609.19587

    红队测试 Auto Mode：用注入与多上下文攻击绕过编码 Agent 阻断监控器

    提出 Auto Mode ++，加固编码 Agent 阻断监视器

    arXiv
    2609.19587
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Opus 5, Opus 4.8, Opus 4.7 and others8

    Abstract系统评估生产阻断监视器在面对持续恶意智能体时的表现，提出加固防御并指明多上下文防御挑战。

    Paper analysisRead analysis
  12. InterpretabilityarXiv 2606.05396

    研究用消融移除代码 LLM 拒答方向，分离拒答与生成能力

    用 abliteration 消融代码 LLM 拒答方向，分离拒答与漏洞注入能力

    arXiv
    2606.05396
    Published
    Layer
    Model layer
    Setting
    coding agent
    Evaluated models
    Qwen2.5-Coder-Instruct 3B, Qwen2.5-Coder-Instruct 7B, Qwen2.5-Coder-Instruct 14B
    Abstract拒答可用低秩编辑去掉，CWE-89 注入是否成功仍由模型规模决定。

    这项初步可行性研究用 abliteration 把代码 LLM 的拒答和 CWE 注入能力拆开：意愿可以被低秩编辑解锁，注入是否成功仍看参数规模。

    Paper analysisRead analysis
  14. Evaluation and benchmarksarXiv 2605.19722

    研究用 30 项漏洞分析任务评测自主安全智能体的安全对齐效果

    评测自主安全智能体在授权漏洞分析中的对齐拒答效应

    arXiv
    2605.19722
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Gemma 4 31B official instruction-tuned model, TrevorJS Gemma 4 31B uncensored GGUF, Gemma 4 26B A4B official instruction-tuned model and others8
    Abstract轨迹级评测把拒答、接地、工具接口分开。

    这项工作把安全对齐效应测成自主安全智能体的轨迹属性，而不是单轮拒答率。

    Paper analysisRead analysis
  15. Evaluation and benchmarksarXiv 2605.16626

    Sleight-Bench: 20 of 40 attacks were never detected by Opus 4.6 monitoring.

    构建 SLEIGHT-Bench 评测代码智能体监控对概念盲点规避的捕获

    arXiv
    2605.16626
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Claude Opus 4.6, Claude Opus 4.7, Claude Sonnet 4.6 and others6

    Abstract评估针对监控盲点的规避基准，发现前沿监控器漏报率高，算力与脚手架虽有提升但存在类别局限。

    Paper analysisRead analysis
  17. Evaluation and benchmarksarXiv 2510.09462

    研究者用提示注入攻击监控模型，绕过 AI 控制协议

    提出针对可信监控器的自适应提示注入，绕过 AI 控制协议

    arXiv
    2510.09462
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    GPT-4o, GPT-4, GPT-4.1 and others4

    Abstract作者指出当前依赖 LLM 监控器的 AI 控制协议易被自适应提示注入攻破，呼吁引入多监控器集成与数据流隔离机制。

    Paper analysisRead analysis
You have reached the end