Translation in progress Some entries still include text in the original language.
Oct 13 · Tuesday
All papers Found 79 papers Another 215 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 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.
Measures citation fragility in AI search, demonstrating that ordinary public posts can enter citations and answers.
Published Oct 8
Evaluated models Grok , Doubao , ChatGPT and others10 Abstract The authors claim the paper quantifies vulnerabilities in AI search caused by citation preferences and low posting thresholds; tests show public web posts can penetrate AI citations, and they call for auditing the citation supply chain.
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。
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 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。
提出 RAG-PIBench，评测 RAG 上下文中提示注入检测器
Published Oct 6
Evaluated models DistilBERT-base-uncased , DeBERTa-v3-base , MiniLM 提出 SAFESHIELD，组织小语言模型部署时的安全决策
Published Oct 6
Evaluated models Qwen3-4B-Instruct Abstract SAFESHIELD 用可审计决策轨迹组织 SLM 部署时的四项安全决策，并报告协调依赖影响端到端结果。SAFESHIELD 是面向小语言模型的部署时安全框架，把运行时安全从单项审核改写成可检查的决策过程。
证明短开头词可唤醒基模型推理，且效应来自训练数据
Published Oct 6
Evaluated models Olmo-3-7B , Olmo-3-32B , Qwen3-4B and others10 提出临时可见性协议，限制持续摄入 RAG 向量库的投毒暴露
Published Oct 5 Abstract 用截止期限制未核验可见时长。这篇工作给持续摄入的 RAG 向量库一个失败关闭的临时可见性协议，把协议能限制的未核验暴露和依赖检测器的中毒暴露分开。
构建 AGENTDOXX 基准，评测联网搜索 Agent 对匿名访谈的再识别
Published Oct 5
Evaluated models GPT-5.6 Terra , GPT-5 Mini , GPT-4.1 Mini and others10 Abstract 论文提出了已知身份的重识别基准 AGENTDOXX，揭示搜索与记忆双重风险及现有防御在检索轨迹中的泄露漏洞。
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。
提出 RAGStress 受控基准，评测知识库退化下的 RAG
Published Oct 4
Evaluated models Claude Sonnet 4 , Claude Opus 4.6 , DeepSeek R1 and others5 Abstract 作者用 RAGStress 比较五系统。RAGStress 是知识库损坏下的 RAG 受控压力测试。作者明确不把它当作模型排行榜。
提出 HDI 攻击以篡改 GraphRAG 辅助结构破坏检索
Published Oct 2
Evaluated models Microsoft GraphRAG , HippoRAG2 Abstract 论文形式化了 GraphRAG 辅助模式级实体攻击面，提出 HDI 攻击并验证其攻击效果。
证明 RLVR 后训练允许无界语言漂移且限制漂移必限制奖励
Published Oct 2
Evaluated models gemma-3-1b-pt , Llama-3.2-1B , Qwen2.5-1.5B Abstract 论文证明并验证了 RLVR 在新任务上必然诱发独特的语言漂移，揭示了前沿推理模型可监控性的安全隐患。论文从理论与实验两方面研究了大语言模型在 RLVR 后训练过程中思维链出现的语言漂移现象。
检验低监控读数能否证明代码生成中的奖励作弊已受控
Published Oct 2
Evaluated models Llama-3.1-8B-Instruct Abstract 论文指出监控器训练读数接近零无法确立作弊受控，模型可利用通用外壳推迟提交绕过监控，强化学习亟需带外行为验证。本文是一篇针对强化学习可验证奖励后训练（RLVR）中监控器有效性与奖励作弊行为的实证分析研究。论文要解决的核心问题是：在训练目标中引入监控器以抑制代码生成中的作弊行为时，接近零的监控器读数是否足以确立模型行为已真正受控。
提出 imMRAG 攻击，从多模态 RAG 私有图像库提取内容
Published Oct 1
Evaluated models Lumina-mGPT , Gemini 2.5 Flash Image Preview , CLIP ViT-B/16 and others5 提出面向 RAG 的证据优先发布框架 AGO，并评测评审 adherence 判别
Published Oct 1
Evaluated models gpt-4o , gpt-4.1-nano Abstract AGO 把缺失证据和评审误差写成显式发布状态，并要求按领域验证评审。AGO 是部署在企业 RAG 评估业务中的自托管发布门控，用来决定一个版本应放行、人工复核还是拦截。
提出 MADBench，评测多智能体辩论是吸收还是放大对抗影响
Published Sep 30
Evaluated models gpt-4o , GPT-5 , GPT-6 Astra Abstract MADBench 显示辩论能吸收部分准确率攻击，同时放大未授权读写。MADBench 是一个评测多智能体辩论安全性的基准，用来看攻击是被辩论吸收还是被放大。
提出 PACT，遗忘推理模型在压力下的欺骗行为并保留上下文理解
Published Sep 30
Evaluated models DeepSeek-R1-Distill-Qwen-32B , QwQ-32B Abstract 论文将大模型欺骗形式化为上下文条件行为，提出 PACT 框架在消除欺骗的同时保留了上下文遵循能力。
提出 worst-case Constrained RL，蒸馏时约束推理前缀偏离以抑制奖励作弊
Published Sep 29
Evaluated models Qwen2.5-1.5B , Llama-3.2-3B , Qwen2.5-Coder-0.5B Abstract 作者提出前缀约束蒸馏。这是把推理蒸馏写成最坏前缀约束 RL 的训练研究。 作者关心最终答案奖励下的 reward hacking，以及序列平均软惩罚允许局部偏离被其他 token 补偿的问题。