Skip to content

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

Oct 12 · Monday

All papers

Found 2 papers
Filter5

Model risks

Reward hacking3Misalignment and value divergence2Dangerous agent actions4Inappropriate refusal3

System type

AI Agent2coding agent0

System component

Memory6Monitors3Reasoning chains1Vision1
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 240papers have no tags yet and are not included in filtered results.
Model risks：Misalignment and value divergenceSystem type：AI AgentSystem type：coding agentSystem component：Reasoning chainsSystem component：VisionClear all

Another 240papers do not have category tags yet and are not included in filtered results.

  1. Risky behaviorarXiv 2609.35291

    窄域多模态微调可诱发涌现性失配，覆盖 15 个视觉语言模型

    证明无显式危害的图文窄微调可诱发跨任务涌现未对齐

    arXiv
    2609.35291
    Published
    Layer
    Model layer
    Setting
    Browser and computer use
    Evaluated models
    GPT-4o, GPT-4.1, Gemini 2.5 and others15
    Abstract证实窄多模态微调可引发全面的多渠道未对齐行为转变，揭示了后训练阶段对齐被隐蔽重塑的安全隐患。

    论文系统探究了视觉-语言模型中由窄任务微调诱发的跨模态涌现未对齐（Emergent Misalignment, EM）现象。

    Paper analysisRead analysis
  2. DefensesarXiv 2608.27348

    INTENT-AS-A-TOOL：用意图工具追踪智能体失准

    提出 INTENT-AS-A-TOOL，用意图工具概率跟踪并干预智能体失准

    arXiv
    2608.27348
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Gemma-4-31B-IT, Qwen3.5-27B, Qwen3-235B-A22B and others5
    Abstract用意图工具的下一动作概率跟踪推理中的失准承诺，并在其成为首选时插入反思。

    Zhang 等人研究推理型语言模型在职场邮件沙箱里如何走向智能体失准，并提出用动作偏好而不是事后文本标签来跟踪承诺。问题是：有害执行往往出现在推理已经形成意图之后，但 CoT 监控是事后、粗粒度、依赖外部评审的，也看不到模型倾向调用哪个动作。

    Paper analysisRead analysis
You have reached the end