Translation in progress Some entries still include text in the original language.
Oct 12 · Monday
All papers Found 18 papers Another 232 papers do not have category tags yet and are not included in filtered results.
“
About this collection Collection in progress We are building a large AI safety paper collection to help researchers find work related to their interests. Papers are collected automatically andfiltered only for relevance to AI safety , without judging quality, novelty or reading value. The default “Annotated first” shows papers with tags or published paper analyses first. Each group is ordered by original publication date, newest first, with undated papers last. “Newest first” uses publication dates only. This order is not a quality ranking.
One-line summaries, tags and paper analyses are prepared by AI from the papers. Language models may have systematic biases and produce hallucinations.We plan to carry out human annotation and review to reduce errors in this content.
An ecological model is established for a population of mismatched agents, showing that collaboration produces a critical takeoff scale.
Published Oct 9 Abstract The 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. .
测量事实核查评测混入的时序证据与参数泄漏
Published Oct 4
Evaluated models Qwen3-14B , phi-4 Abstract 两条后见通道会抬高事实核查分数。作者量的是 LLM 事实核查分数里，有多少来自 claim 当时还不可知的信息。通道有两条：预训练记住的结局，以及检索到的、写于 claim 之后的文档。
分析文本水印在 RAG 中诱发事实幻觉的机制并提出干预
Published Oct 4
Evaluated models LLaMA 3.1 8B Instruct , Mistral-7B-Instruct-v0.3 , Qwen2.5-14B-Instruct Abstract 论文揭示了生产环境文本水印导致模型事实退化的失效模式，分析了解码机制并提出针对性干预。论文研究生产级文本水印对大语言模型事实准确性的影响，分析了水印诱发幻觉的失效模式并评估了针对性缓解方案。
量化异构模型在分歧分类上的同伴影响与回弹
Published Oct 2
Evaluated models gemma-3-4b , gemma-3-27b , llama-3.1-8b and others7 Abstract 一次分歧交换就能大幅改判，且组合行为不能从个体属性外推。这篇工作测量异构语言模型智能体在共同分类任务上的单次同伴影响，而不是评多智能体系统的整体基准分。
提出奖励驱动多智能体系统 MEA，生成可扰动验证的忠实解释
Published Oct 2
Evaluated models Qwen3.6-35B-A3B , Qwen3.5-4B , Qwen3-VL-30B-A3B-Instruct and others12 Abstract MEA 用忠实性奖励训练双智能体，把工具输出收成可扰动核对的解释。MEA 是一个奖励驱动的双智能体，用来给黑盒分类器的单次预测写自然语言解释，并检查解释能否被扰动后的模型行为支持。
揭示安全路由评测在分布偏移下的虚假下限与选择偏差
Published Oct 1
Evaluated models claude-3-5-sonnet , claude-3-opus , claude-sonnet-4-6 and others9 Abstract 论文揭示安全路由评估在分布偏移下的虚假下限，指出防御重心应转向外部动作级强制。
用含压抑率的意见动力学解释多智能体讨论何时优于投票
Published Sep 30
Evaluated models glm-flash , seed , ling and others9 Abstract 讨论增益由压抑率相对 c∗=γ/(γ+a) 的裕度决定，关闭推理或减少压抑可增大该增益。作者用一个只含四类行为的意见动力学，解释 LLM 团队讨论何时优于多数投票、何时停在错误共识。
提出 VStress 回放审计，并按条件边际信息分配重复验证预算
Published Sep 29 Abstract VStress 冻结后再评分，VStress-CA 按条件信息分配调用，相关共享时增益消失。VStress 是重复二元验证器的回放审计，VStress-CA 把未查询通道的条件边际信息变成固定预算下的停止与分配。
提出 REGEXROUTE，用 SAE 引导的正则特征做 LLM 路由
Published Sep 28 分析零训练 LLM+OVOD 流水线中 CAAP 与 SNAP 分离的来源
Published Sep 26
Evaluated models Gemma-3 4B , YOLO-World X , Qwen2.5-VL 7B and others4 分析稀疏重叠下 LLM judge 与人类一致性的验证误差并规划重叠分配
Published Sep 26
Evaluated models Gemini Flash , Gemini Pro , GPT-4o-mini and others10 Abstract 稀疏重叠主导 judge 验证的错误决策。这篇工作研究标注重叠很稀时，怎样验证 LLM judge 是否足以替代人类，并规划重叠该留多少、怎么分配。 。部署前的标准做法是在校准集上比较 judge 与人类的一致性。
归纳开放式医学长答案检索核验的失败分类
Published Sep 25
Evaluated models GPT-5.4 , Mistral Small 3 (24B-Instruct-2501) , Mistral Small 4 (119B-2603) and others8 Abstract 作者把开放式低召回归为证据是否在库、能否被定位，以及与声明是否对齐。这篇工作诊断开放式医学事实核验的 retrieve-then-verify 管线断在哪里，而不是提出新的核验器。
审计 Active Inference Agent 上 LLM 解释器的监督叙述失败
Published Sep 20
Evaluated models GPT-4o , Claude-3-Opus , Gemini Abstract 三组触发下解释流畅但错误，所提缓解都未被评估。作者审计的是挂在智能体上的运行时解释器，而不是智能体本身。操作员读到的是信念与动作的自然语言叙述。作者要问哪些可复现触发会让这段叙述失败，以及失败痕迹是什么样。
提出 Trace2ATT&CK，用溯源图把内核遥测映射到 ATT&CK
Published Sep 11
Evaluated models gpt-oss-20b , gpt-oss-120b , gemma-4-31b-it and others7 Abstract Trace2ATT&CK 把 eBPF 溯源命令图交给本地 LLM/RAG，在 Atomic Red Team 上映射 ATT&CK。Trace2ATT&CK 是一条从 Linux 内核遥测到 MITRE ATT&CK 排序候选的端到端流水线，面向已经划定的攻击窗口，而不是异常检测或 CTI 报告。
刻画单迹监控对跨执行超性质的监督缺口
Published Sep 7
Evaluated models Llama-3.1-8B , Mistral-24B , GPT-4o-mini and others9 Abstract 单迹监督的上界由 TV 决定。论文测量单迹安全监控相对 2-safety 超性质 能看见多少：跨租户泄漏、觉察到被评估、策略性表现不佳都由两次执行之差定义，一条轨迹装不下判定。
分析多智能体在压力协作中是否涌现外部难监控的语言
Published Sep 2
Evaluated models GPT 5.4 , Sonnet 4.6 , Opus 4.7 and others5 Abstract 专有模型在预算压力与轮间协商下发展出可产、可传递的非英语代码。GlossoGen 中的 SaveVeyru 用来观察预训练 LLM 智能体在必须通信时会不会离开英语。
用 Ising 式耦合规则预测语言模型智能体的集体意见演化
Published Aug 17
Evaluated models GPT-4o-mini , Gemma-3n-E4B-it , Qwen3.5-9B and others6 Abstract 统计力学三耦合模型从初始意见预测智能体社群轨迹，并解释 conviction、共识与客观题求真。定位 ：用统计力学描述语言模型智能体社群的意见动力学，并从初始意见预测后续轨迹。
Arguing that safety alignment continuously disappears in self-evolving AI societies
Published Feb 10
Evaluated models Qwen3-8B Abstract The authors claim that closed self-evolution makes safety alignment unsustainable.The authors discuss whether a closed self-evolving agent society can simultaneously maintain safety alignment.
You have reached the end