Skip to content

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

Oct 13 · Tuesday

All papers

Found 8 papers
Filter4

Model risks

Deception and scheming12Reward hacking13Misalignment and value divergence4Dangerous agent actions24Dangerous capabilities9Hallucination and fabrication4

Impact

Unauthorized actions3

System type

Models and APIs59LLM applications9AI Agent35coding agent7science agent3Browser and computer use1Multi-agent systems11

Layer

Model layer3Application layer5
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 213papers have no tags yet and are not included in filtered results.
Model risks：Misalignment and value divergenceModel risks：Hallucination and fabricationSystem type：coding agentSystem type：Browser and computer useClear all

Another 213papers do not have category tags yet and are not included in filtered results.

  2. 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
  3. 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
  4. OtherarXiv 2609.12039

    现实才是最终验证者：智能体软件工程的两大关键缺口

    提出双缺口框架，统一智能体软件工程中的奖励投机与幻觉

    arXiv
    2609.12039
    Published
    Layer
    Application layer
    Setting
    coding agent
    Abstract双缺口解释内环为何会放行部署后不可接受的实现，外环用现实证据持续修订 R、M 与 E。

    Reality Is the Final Verifier 用 two-gap framework 解释智能体软件工程的主要失败，并主张用外层保证环持续收窄缺口，而不是试图在部署前证明缺口已关闭。

    Paper analysisRead analysis
  6. AttacksarXiv 2607.12340

    Research: LLM agent skill-name hallucination rates reach up to 62%, which can be exploited via squatting for supply-chain attacks

    提出技能名称抢注攻击，利用 Agent 虚构技能名实施供应链劫持

    arXiv
    2607.12340
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    GLM-4.5-Air, GPT-5.4-mini, Claude Sonnet 4.6 and others12
    Abstract研究证实智能体虚构技能名具有高发生率与可预测性，现有防御难以兼顾安全与效用，亟需注册表结构性防护。

    本研究系统揭示了大模型智能体技能推荐场景下的虚构技能名称安全风险，将其形式化为一种新型供应链抢注攻击。

    Paper analysisRead analysis
  8. Evaluation and benchmarksarXiv 2605.17062

    研究复现五个前沿模型的包名幻觉：跨模型共同幻觉 127 个包名，53 个仍可被注册

    复评前沿模型代码生成中的包名幻觉与抢注攻击面

    arXiv
    2605.17062
    Published
    Layer
    Model layer
    Setting
    coding agent
    Evaluated models
    Claude Sonnet 4.6, Claude Haiku 4.5, GPT-5.4-mini and others5

    Abstract评估显示前沿模型包幻觉率收窄但威胁仍在，作者识别出跨模型通用攻击面并呼吁联合防御。

    Paper analysisRead analysis
You have reached the end