Skip to content

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

Oct 12 · Monday

All papers

Found 56 papers
Filter4

Attack type

Jailbreaks3Prompt injection5

Model risks

Deception and scheming3Reward hacking5Dangerous agent actions6Dangerous capabilities8Inappropriate refusal4

Defense type

Inference-time intervention3Permissions and approval3Monitoring and auditing4

Impact

Harmful content6Unauthorized actions10Output manipulation8Availability and reliability6Oversight failure4

System type

Models and APIs407LLM applications127AI Agent609coding agent123science agent30web agent34Browser and computer use19Multi-agent systems78Embodied systems and robotics27

Attack surface / entry point

User input7Images, audio and video3

Layer

Prompt layer6Model layer10Application layer46Training and data layer16

System component

Guardrails and judges4Monitors5Permissions and sandboxing3Vision7Multimodal8
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 261papers have no tags yet and are not included in filtered results.
System type：science agentSystem type：Embodied systems and roboticsLayer：Model layerLayer：Application layerClear all

Another 261papers do not have category tags yet and are not included in filtered results.

  1. OtherarXiv 2610.13479

    A local agentic AI framework generates edge-case scenarios for safety assurance

    Proposes a local dual-model closed loop to generate edge scenarios for autonomous system safety assurance

    arXiv
    2610.13479
    Published
    Layer
    Application layer
    Setting
    LLM applications, Embodied systems and robotics
    Evaluated models
    PPO
    AbstractLocal closed-loop generation of JSON edge cases.

    The authors built a local agentic loop that runs on a laptop to generate edge cases for safety arguments of autonomous systems, rather than releasing a new public test set.

    Paper analysisRead analysis
  3. OtherarXiv 2610.13372

    Toward Safe World Models: Risks, Defenses, and Evaluation

    A survey of risks, defenses, and evaluation in the six-stage lifecycle of world models

    arXiv
    2610.13372
    Published
    Layer
    Application layer
    Setting
    Embodied systems and robotics
    AbstractSafety is written as a property of the prediction-decision closed loop; claims must be linked to observed consequences and versions.

    World model safety is organized by the nature of the prediction-decision loop, not by isolated prediction scores.

    Paper analysisRead analysis
  6. DefensesarXiv 2610.11754

    仅凭输出审查无法验证什么：面向研究智能体的研究契约

    提出研究合同，把已批准选择绑定到执行证据并由检查器核验

    arXiv
    2610.11754
    Published
    Layer
    Application layer
    Setting
    science agent
    Evaluated models
    OpenAI gpt-5.6-sol, OpenAI gpt-5.6-terra, OpenAI gpt-5.6-luna and others18
    Abstract研究合同暴露四类输出单独建不立的关系。

    研究合同用来核验计算科学研究智能体是否仍在已批准研究之内，不判断科学是否为真。。

    Paper analysisRead analysis
  8. Analysis and theoryarXiv 2610.09517

    审计发现 CLIP 安全分数主要反映画面与提示词场景的相似度

    开环审计冻结视觉语言安全分，发现其反映场景模板而非危险感知

    arXiv
    2610.09517
    Published
    Layer
    Model layer
    Setting
    Embodied systems and robotics
    Evaluated models
    CLIP ViT-B/32, CLIP ViT-L/14, Qwen2-VL-7B
    Abstract论文审计了冻结 VLM 安全评分机制，指出其反映场景模板而非危险感知，为安全强化学习评估提供了严谨基准。

    本文是一项针对安全强化学习中冻结视觉语言模型（VLM）安全评分机制的实证审计研究。

    Paper analysisRead analysis
  10. OtherarXiv 2610.08917

    CARE：为视觉-语言-动作模型推理加速提供认证式选择

    提出 CARE，离线认证 VLA 加速配置的回合级失败风险并选出最快合格候选

    arXiv
    2610.08917
    Published
    Layer
    Model layer
    Setting
    Embodied systems and robotics
    Evaluated models
    OpenVLA-OFT, π0.5, π0 and others5
    AbstractCARE 以配对回合认证加速诱发失败，在预算内选最快候选，不足则保留参考。

    CARE 在部署前从加速配置中选出最快、且加速诱发失败风险能被有限样本认证为低于用户预算的候选。证据不足就保留全计算参考策略。

    Paper analysisRead analysis
  14. AttacksarXiv 2610.06083

    研究者提出面向深度强化学习的轨迹级可迁移黑盒对抗攻击

    提出轨迹级可迁移对抗攻击，降低深度强化学习策略的期望回报

    arXiv
    2610.06083
    Published
    Layer
    Model layer
    Setting
    Embodied systems and robotics
    Evaluated models
    DQN, Double DQN (DDQN)
    Abstract轨迹级后退视界攻击在 CartPole 上比逐步 FGSM 类和 Sign 噪声更压低回报。

    这项工作研究 DRL 上的迁移式黑盒观测攻击：受害参数不可访问时，在白盒代理上造扰动再迁移。

    Paper analysisRead analysis
  15. OtherarXiv 2610.05689

    研究者提出用可逆神经网络为 AI 控制系统寻找可证明的前向不变集

    提出用可逆网络为给定 AI 控制器寻找可证明前向不变集

    arXiv
    2610.05689
    Published
    Layer
    Model layer
    Setting
    Embodied systems and robotics
    Evaluated models
    Pendulum AI-controller, Cart pole AI-controller, PVTOL AI-controller
    Abstract用潜空间超矩形加解析验证，为事先给定的 AI 控制器寻找 FIS。

    本文处理的是认证场景：AI 控制器事先给定，不能为了构造 Lyapunov 函数去改控制器，却仍要给出可解析核验的安全依据。

    Paper analysisRead analysis
  16. DefensesarXiv 2610.05166

    安全动作不等于可行动作：VICS-G 为 VLA 策略做可行未来解码

    提出 VICS-G 可行未来解码，重排冻结 VLA 策略的动作候选

    arXiv
    2610.05166
    Published
    Layer
    Model layer
    Setting
    Embodied systems and robotics
    Evaluated models
    SafeVLA (task-optimized / base), SafeVLA (safety-aligned / safe)
    Abstract作者用可行未来解码重排冻结 VLA 的下一步，并称 VICS-G 降低观测代价且完成率接近策略采样。

    训练无关解码器 VICS 在冻结 VLA 的推理阶段重排动作块，使当前选择带上该动作仍可能留下的安全完成质量。

    Paper analysisRead analysis
  17. Evaluation and benchmarksarXiv 2610.05140

    AutoSciBench：自动生成科学智能体评测基准的框架

    提出 AutoSciBench，循环生成科学智能体评测任务

    arXiv
    2610.05140
    Published
    Layer
    Application layer
    Setting
    science agent
    Evaluated models
    Claude Opus 5, Claude Opus 4.8, Claude Opus 4.7 and others10
    AbstractAutoSciBench 用概念、配方和求解反馈自动构造并加难科学智能体基准。

    AutoSciBench 研究科学智能体基准能否自动构造，并随 solver 行为迭代改编。

    Paper analysisRead analysis
  18. Evaluation and benchmarksarXiv 2610.04366

    CrashSim 用真实事故先验生成碰撞场景并构建 nuCrash 数据集

    提出 CrashSim 用真实事故先验引导扩散仿真并构建 nuCrash 评测自动驾驶规划器

    arXiv
    2610.04366
    Published
    Layer
    Application layer
    Setting
    Embodied systems and robotics
    Evaluated models
    Lane-graph planner, PDM-closed planner, IL planner and others5
    AbstractCrashSim 迁入真实事故先验生成场景。

    CrashSim 用检索到的真实事故先验引导扩散式多车仿真，为自动驾驶安全评测生成碰撞与近碰撞场景。。

    Paper analysisRead analysis
  19. Evaluation and benchmarksarXiv 2610.04184

    EvalResearchBench：AI 智能体自主设计评测基准

    提出 EvalResearchBench，评测智能体自建评测对目标基准的排序一致性

    arXiv
    2610.04184
    Published
    Layer
    Application layer
    Setting
    science agent
    Evaluated models
    GPT-6 Astra, GPT-5.6 Sol, GLM-5.3 and others9
    AbstractERB 比较有预算的评测设计，分数复现与排序并不由同一设计同时领先。

    ERB 用来测一件事：智能体在固定预算内做出的可执行评测，能否复现既有基准对候选模型的分数和排序。

    Paper analysisRead analysis
Load more