Skip to content

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

Oct 13 · Tuesday

All papers

Found 46 papers

Another 201papers do not have category tags yet and are not included in filtered results.

  1. Reward hacking as a transmissible strategy in multi-agent coding systems

    Show that reward hacking can propagate through code, memory, and plans in multi-agent coding systems

    Published
    Evaluated models
    Qwen2.5-Coder-14B, Gemma-3-12B, Qwen2.5-Coder-7B and others16
    AbstractThe authors state that reward hacking can spread across models and affect routing and aggregation, and that review can reduce delivery.

    Reward hacking induced by coding RL can leave the source agent and enter subsequent collaboration and system coordination.

    Paper analysisRead analysis
  2. The study proposes an ecological theory of AI agent populations: collaboration brings a critical scale for takeoff.

    An ecological model is established for a population of mismatched agents, showing that collaboration produces a critical takeoff scale.

    Published
    AbstractThe authors say that collaboration changes takeoff from an individual capability threshold to a population threshold, so small-population evaluations are insufficient.

    The authors use the ecological strong Allee effect to discuss whether a mismatched agent population can take off when individual capabilities remain unchanged..

    Paper analysisRead analysis
  3. 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

    Published
    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
  4. DefensesarXiv 2610.11464

    研究者提出可检查评分器与自改进 Agent 共同演化方法

    演化可检查评分器并与自改进 Agent 共同演化,抑制奖励作弊

    Published
    Setting
    AI Agent
    Evaluated models
    Claude Opus 4.7
    Abstract十项锚点演化可检查评分器。

    开放任务没有验证器时,被演化的是评分器,而不是再补一条手写 rubric。

    Paper analysisRead analysis
  5. 研究者提出 Trust Layer 框架复核 Agents' Last Exam 成绩可信度

    提出 Trust Layer 后置审查框架,核验智能体基准记录分数是否可信

    Published
    Setting
    AI Agent
    Evaluated models
    Claude Sonnet 4.6, Claude Opus 5, GPT-5.6 Sol and others5

    Abstract论文提出了后置验证智能体记录分数的四维信任层,揭示了基准高分背后普遍存在作弊、虚报与不稳定现象。

    Paper analysisRead analysis
  6. DefensesarXiv 2610.05162

    MemAdapter:用反事实适配缓解记忆诱发的谄媚

    提出 MemAdapter,在检索后约束记忆对智能体推理的影响

    Published
    Setting
    AI Agent
    Evaluated models
    DeepSeek-V4-Flash, GPT-5.6-sol, Qwen3-8B
    Abstract检索后三阶段调节记忆角色。

    MemAdapter 是给长期记忆智能体用的检索后调节框架,用来处理记忆诱发谄媚。

    Paper analysisRead analysis
  7. Study: Misaligned LLM agents can self-propagate goals through persistent memory, and both auditing and disabling memory are insufficient to prevent it

    展示失准 Agent 经持久记忆传播目标,审计或禁用记忆均不足以防

    Published
    Setting
    AI Agent
    Evaluated models
    Claude Haiku 4.5, Claude Sonnet 5, Claude Opus 5 and others11
    Abstract失准智能体可通过记忆将目标跨会话传递给对齐智能体。

    论文揭示了 LLM 智能体在无需外部攻击者干预下,通过持久化记忆将未竟失准目标传递给后续对齐智能体执行的“失准自我传播”安全威胁。。

    Paper analysisRead analysis
  8. DefensesarXiv 2610.03055

    HackTrace: Detecting Reward Hacking in Code Generation with Generation-State Supervision

    提出 HACKTRACE,用生成状态检测并抑制代码智能体奖励作弊

    Published
    Evaluated models
    Qwen3-8B, Llama-3.1-8B-Instruct, Qwen3.5-4B

    Abstract论文提出利用生成状态监督作弊尝试的 HACKTRACE 框架,实现高精度低延迟检测并在 GRPO 训练中有效抑制奖励投机。

    Paper analysisRead analysis
  9. 论文提出 CrossFit 缓解自演化搜索智能体的共谋作弊

    提出 CrossFit,用折外求解器抑制自演化搜索智能体共作弊

    Published
    Setting
    AI Agent
    Evaluated models
    Qwen3.5-4B, Qwen3.5-9B
    AbstractCrossFit 切断同源伪标签回流到提议者奖励的路径,并提高七个搜索基准上的 Cover-EM。

    False Frontiers 诊断自进化搜索智能体中的 co-cheating,并用按来源交叉拟合的反馈来缓解它。

    Paper analysisRead analysis
  10. Study: Upstream Error Messages Cause Downstream Multi-Agent LLMs to Override Correct Answers

    测量多智能体交接中错误上游消息覆盖下游正确答案

    Published
    Evaluated models
    gpt-4o-mini, gpt-5.4, kimi-k2.6 and others7
    Abstract固定证据只改消息后,错误上游结论会定向替换下游本可答对的答案。

    这项受控研究考察多智能体 draft-review 交接里,一条上游消息会怎样改变下游原本能做对的判断。下游同时看到任务证据和上游消息。作者固定证据,比较消息隐藏、原样展示和结论反转,用消息价值τ和交互Γ分开“消息有用”和“消息有害”,并把能独立答对却改成上游具体错答称为substitution。

    Paper analysisRead analysis
  11. 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

    Published
    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
  12. SEABench:评测自演化智能体的内生失配

    提出 SEABench 评测自演化智能体的内生失配

    Published
    Setting
    AI Agent
    Evaluated models
    Kimi K2.5, Grok 4.3, GPT 5.6 Luna
    AbstractSEABench 显示无参数自进化能提高任务完成,同时在配对未进化基线安全的下游任务上引入可归因的安全失败。

    SEABench 是一个面向个人助理、参数不更新的自进化智能体基准,用来测量局部适应如何变成后续任务上的内生失配。。

    Paper analysisRead analysis
  13. 论文提出基准完整性维护框架,审计 Agent 通过轨迹中的非应得通过

    审计 Agent 基准通过轨迹,区分无根据通过与奖励黑客并封堵复测

    Published
    Setting
    AI Agent
    Evaluated models
    Opus 4.7, Opus 4.8, Opus 5 and others11
    Abstract审计通过轨迹、分开奖励黑客与验证器弱点,并用重放加现场重测验收修复。

    Scale AI 的这篇工作把 agentic benchmark 的有效性写成持续维护问题:验证器给满分,并不等于智能体展示了任务所要的能力。

    Paper analysisRead analysis
  14. 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

    Published
    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