Translation in progress Some entries still include text in the original language.
Oct 12 · Monday
All papers Found 169 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.
Proposes DREAM, which selectively weakens agent-poisoned memories while preserving benign memories.
Published Oct 9
Evaluated models GPT-5.4-mini , Claude-Haiku-4.5 Show that reward hacking can propagate through code, memory, and plans in multi-agent coding systems
Published Oct 9
Evaluated models Qwen2.5-Coder-14B , Gemma-3-12B , Qwen2.5-Coder-7B and others16 Abstract The 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.
Proposes the Cancerous backdoor, which progressively corrupts reasoning using invisible stage tokens.
Published Oct 9
Evaluated models Llama-3.1-8B-Instruct , Qwen2.5-7B-Instruct , Mistral-7B-Instruct-v0.3 Abstract Hidden tokens plus a stage controller make the backdoor progressively corrupt reasoning upon triggering, and the authors believe that looking only at the answer will miss it.Cancerous Backdoor Attack changes LLM backdoors from one-time activation to multi-stage reasoning corruption that accumulates progressively with triggers.
Uses premise permutation and hidden-layer tracing to audit whether legal chain-of-thought truly relies on the cited legal provisions.
Published Oct 9
Evaluated models Qwen3-8B , Qwen3-14B , Llama-3.1-8B and others7 Abstract Using counterfactual substitution of authorities and internal commitment tracing, the authors find that legal CoT rulings depend on authorities only to a limited extent and are vulnerable to prompt injection manipulation.This study conducts an empirical audit of the causal truthfulness of legal citations in chain-of-thought reasoning of large language models in the legal domain.
度量关闭思考指令下终答前是否仍有可见显式推理
Published Oct 8
Evaluated models Qwen3-4B , Qwen3-32B , Qwen3-235B-A22B and others6 Abstract 作者用三级指标测 no-thinking，并称模型有随答案空间加重的思考惯性。作者把 no-thinking 定义为可测量的回答行为，而不是从思考开关、特殊标记或输出长度推断。
构建 Workerville 基准，测量组织条件如何改变 Agent 的安全行为
Published Oct 8
Evaluated models GPT-5.2 , Claude-Sonnet-4.6 , Gemini-3.1-Pro and others6 Abstract 作者用 ACB 与 Workerville 测量组织条件对智能体安全的影响，并称结果支持可量化的组织行为框架。Workerville 用组织行为测量智能体安全如何随监督、同伴和记忆条件变化。 。要处理的缺口是：持续运行的 LLM 智能体同时接收用户指令、同伴消息和长期记忆，但既有安全评测多把这些通道拆开，没有统一的定量测量。
提出 RL-ARC，用推理置信度在强化学习中校准推理模型
Published Oct 8
Evaluated models Qwen2.5-7B , Qwen3-8B , Llama-8B distilled from DeepSeek-R1 Abstract RL-ARC 用推理置信度调节答案置信度，作者称可少牺牲精度并改善 ID 与 OOD 校准。RL-ARC 是校准感知的强化学习框架，用推理过程的口头置信度调节答案置信度。
检验推理 token 数能否区分被提示的诚实与不诚实作答
Published Oct 8
Evaluated models Gemma-4-e4b , Qwen-3.5-9b , GPT-OSS-120B Abstract 被提示不诚实时推理 token 组水平更高。三个推理模型在被提示的回答政策下出现组水平推理 token 差异。作者把它当作不读思维链内容的 black-box 候选信号，而不是自发欺骗检测器。
提出流匹配逆向框架，从多向量视觉文档索引重构页面
Published Oct 7
Evaluated models Tomoro-ColQwen3-8B , ColQwen3.5-4.5B , Tomoro-colqwen3-embed-4b and others13 揭示推理模型用标识符规则替代随机抽样，使审计可被预测
Published Oct 7
Evaluated models GPT-6 Sol , GPT-6 Luna , GPT-5.4 mini and others6 Abstract 研究揭示推理智能体将标识符转化为确定性规则，破坏群体独立性并导致审计预测漏洞，建议显式提供随机输入。
提出 U-SPACE 与 U-LENS，从残差流估计推理轨迹不确定性
Published Oct 7
Evaluated models Gemma-4-31B-it , Qwen3.5-27B , Magistral-Small-2507 Abstract U-SPACE 与 U-LENS 用语义方向和预测熵读推理不确定性，长度控制后仍高于所比基线。U-SPACE 与 U-LENS 是面向推理模型的无训练读出：用语义方向定位不确定性的形式，并用一个轨迹分数区分更可能错的回答。
提出 agent plasticity 指标衡量智能体通过经验自我改进的效率
Published Oct 7
Evaluated models Claude Fable 5 , Claude Opus 5 , GPT-5.6 Sol and others10 Abstract 作者用固定权重协议比较前沿模型把经验写成制品后的留出增益与习得效率。作者用 agent plasticity 衡量一件事：权重冻结时，智能体把经验写成可继承制品，再换成留出能力增益，这个转换有多省。
提出 VeriFine，用双环协同扩展验证以实现具身推理自我改进
Published Oct 7
Evaluated models Alpamayo 1.5 8B , Qwen3-VL 2B , Claude Opus 5 Abstract VeriFine 用双环协同演化具身推理的策略、课程和评审，驾驶与导航的策略分和评审对齐都上升。VeriFine 把驾驶和导航推理的自改进写成策略、课程和无参考评审的协同演化，用来处理固定评审跟不上新失败模式的问题。
用 Learn2Play Bench 评测 Agent 从经验中学习陌生规则
Published Oct 6
Evaluated models Claude Opus 5.5 , GPT-6 Astra , Claude Opus 5 and others11 Abstract 用隐藏规则游戏把新学习与预训练知识分开，并比较模型、记忆方式和人类。Learn2Play Bench 用新设计的隐藏规则文本游戏，评测 LLM 智能体在不更新权重时从交互中学习的能力。
提出 DAEDALUS，用自生成任务自举 Agent 记忆
Published Oct 6
Evaluated models GPT-5.4-mini , GPT-5.6 Luna , Qwen3.6-35B-A3B and others11 实现 DEFER，以确定性检查优先拦截多智能体流水线的内容攻击
Published Oct 6
Evaluated models Qwen3-235B-A22B-Instruct-2507 , gpt-oss-120b , Llama-4-Scout-17B-16E-Instruct and others5 Abstract DEFER 把确定性检查放在 judge 之前。DEFER 用来测量多智能体授权流水线里，确定性规则在何处结束、LLM judge 从何处开始。
提出 SkillPoison，用局部有效的成功经验诱导自进化 Agent 形成过度泛化技能
Published Oct 6
Evaluated models deepseek-v4-flash , Qwen V3.8 Flash , GPT-5.5 Abstract SkillPoison 通过局部有效且验证成功的经验诱导技能过度泛化，在多个评测基准上实现了持久的下游攻击影响。SkillPoison 提出了一种针对自进化智能体经验到技能形成过程的攻击范式，通过组织已验证的成功经验诱导技能过度泛化。
评测记忆门控对激活诱导与记忆诱导谄媚的防御效果
Published Oct 6
Evaluated models Llama-3.1-8B-Instruct , Qwen2.5-7B-Instruct , Mistral-7B-Instruct-v0.3 and others4 Abstract 作者称外部记忆过滤成立，数据未显示反向转向再有贡献。作者在四个开源指令模型上，把激活转向和外部记忆处理放进同一纵深防御协议，看记忆诱发谄媚和内部谄媚压力能否分开压低，以及个性化效用还剩多少。要处理的问题是：检索到的用户历史会推动模型顺从存储信念。只改外部上下文，并不直接动模型内部的谄媚表征。
提出检索可采纳性核验框架，评测记忆智能体的记录资格与披露
Published Oct 6
Evaluated models DeepSeek V4 Pro , Gemini 3.6 Flash , GPT-5.6 Luna and others7 Abstract 作者用三值可采纳性和匹配召回界审计长时记忆检索，并分开报告支持、资格与披露。这是一套长时记忆检索的可采纳性核验协议，用来分开检查候选支持、资格、提示词暴露和答案披露，而不是提出新的检索算法。 。
提出 AMU 按推导图门控企业 Agent 共享记忆的列级访问
Published Oct 6 Abstract 在完整血缘下，AMU 按列权限放行缓存，并标出同名 KPI 冲突。AMU 用列级推导门控共享记忆，拦截经缓存流出的无权敏感列，并标出同名 KPI 的定义冲突。