Translation in progress Some entries still include text in the original language.
Oct 13 · Tuesday
All papers Found 23 papers Another 213 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.
Propose KVeil, which adds perturbations to face videos to prevent unauthorized customization from reproducing identity
Published Oct 9
Evaluated models Wan2.2-TI2V-5B , HunyuanVideo-1.5 , Kling 3.0 and others4 Abstract KVeil adapts LETHE's reference-reading disruption to the challenge's two stages, ranking first on the leaderboard in both Final columns.
提出 OPUR 攻击,用似然梯度优化对抗后缀越狱 LLM 智能体
Published Oct 7
Evaluated models Llama-3.1-8B-Instruct , Ministral-8B-Instruct-2410 提出危害窗口与区间指标,统一评测分解攻击和提示注入下的智能体滥用监视
Published Oct 5
Evaluated models Gemma 4 26B-A4B , Qwen3.6 27B , Qwen3.5 27B and others12 Abstract 统一智能体滥用监视框架,动作视角兼顾两类威胁但危害定位仍存瓶颈。
提出 DART,用去噪表征转移检测多轮智能体攻击并注入提醒
Published Sep 30
Evaluated models Qwen3-8B , Qwen3-14B , Qwen3-32B and others6 提出免训练框架 SED,将有害智能体轨迹蒸馏为可检索安全策略
Published Sep 29
Evaluated models DeepSeek-V4-Flash , GLM-5.2 , Kimi K3 提出 ORBIT 框架,评测多智能体防御对合谋等威胁的效果
Published Sep 27
Evaluated models GPT-5.4 , Claude Sonnet 4.6 , Qwen3-235B and others5 Abstract ORBIT 用六层配置比较多智能体防御,逐动作监控不迁移到合谋。
提出 RLCDAlignBench,零样本检测多种对齐失败
Published Sep 24
Evaluated models Jev (jev-1.13.0) Abstract 一次通用问题已能给多数对齐失败排序,大的误差来自状态和标签。RLCDAlignBench 测量用强化学习做校准决策的 Jev,能否当作对齐失败的零样本检测器。
提出闭环自适应红队框架 CART,评测模型与 Agent 的安全失败
Published Sep 23
Evaluated models Claude Opus 4.8 , GPT-5.5 , Grok 4.3 and others7 Abstract CART 把红队做成可审计的闭环搜索,在三类评测上比静态回放发现更多失败与更高平均风险。CART 把自动红队从一次清单式打分,改成带预算、可追溯的闭环搜索。
系统梳理智能体执行链路中的越狱攻击与防御并比较推理时防御
Published Sep 11
Evaluated models Qwen3.5-4B , Llama-3.1-8B-Instruct Abstract SoK 将越狱重放到智能体流水线,并用两种模型检验回复层防御的边际与代价。Mia 等人把越狱安全从单轮回复扩展到智能体的规划、记忆、工具和智能体间通信,并问既有结论在更强原生对齐下是否还成立。
为冻结模型与目标域自动搜索可部署的 Agent 安全 harness
Published Sep 5
Evaluated models Sonnet 4.6 , GLM-5 , Kimi-K2.5 and others11 提出 SafeEvolve,用轨迹证据协同演化安全 harness 与策略
Published Sep 3
Evaluated models Qwen3.5-4B , Qwen3-4B-Instruct-2507 , Qwen3.5-1.7B and others4 Abstract SafeEvolve 用轨迹证据协同更新安全 harness 和策略,在两个 4B 模型的智能体安全基准上改变安全与效用。SafeEvolve 把已完成的 on-policy 轨迹变成安全 harness 的有界更新,并用两阶段 SFT-RL 把这些更新内化进工具使用策略。
提出 SEAL,在 MoE 共享专家上做 DPO-LoRA 对齐以抵御越狱与权重篡改
Published Sep 2
Evaluated models Qwen1.5-MoE-A2.7B , DeepSeekMoE-16B , GLM-4.7-Flash and others4 Abstract SEAL 把 DPO-LoRA 限制在混合 MoE 共享专家上,SEAL++再保护已有安全神经元方向。SEAL 是面向混合 MoE 的训练时防御:只对齐始终执行的共享专家,使安全不依赖路由器是否把 token 送到安全专家。
训练自博弈红队智能体 GPT-Red,规模化生成提示注入与越狱攻击
Published Jul 29
Evaluated models GPT-5.6 , GPT-5.5 , GPT-5.4 and others8 提出 AHA 自动科研循环,发现生产 Agent 的可复用漏洞概念
Published Jul 13
Evaluated models Minimax-M2.7 , Kimi-K2.6 , Deepseek-V4-Pro and others5 Abstract 论文提出基于可证伪科研循环的 AHA 框架,从生产级智能体中挖掘并沉淀漏洞概念图,提升攻击复用率并指导防御加固。AHA 是一项针对生产级 LLM 智能体自动化红队测试的研究,旨在解决传统方法仅保留具体攻击载荷而缺乏成因解释、导致跨版本复用困难且无法指导修复的问题。
提出 SMT 框架,用模拟审核轨迹越狱函数调用 LLM
Published Jul 1
Evaluated models GPT-4o , GPT-5.4 , Qwen3-Max and others7 提出 Agent MechSuits,对多轮 CLI Agent 做步级检测与子空间转向
Published Jun 22
Evaluated models LLaMA-3.1-8B , Qwen-2.5-7B , Gemma-2-9B Abstract 白盒步级探针加稀疏转向可压低三模型恶意动作,同时提高生成崩溃率。Agent MechSuits 是面向多轮 CLI 智能体的白盒、步级表示防御,并配有可执行轨迹基准 MAS。
提出 CoT Forgery,经用户消息或工具输出注入伪造推理
Published Feb 23
Evaluated models gpt-oss-20b , gpt-oss-120b , o4-mini and others11 Abstract 提示注入被作者解释为角色感知失败:听起来像某角色,就在表示里成为该角色。作者把提示注入解释成角色感知失败:标签划出的权限没有进入模型的内部几何。
提出 LODO 评测恶意提示探针的跨数据集泛化
Published Feb 15
Evaluated models Llama-3.1-8B-Instruct , Gemma-3-27B , Qwen-3.5-4B and others5 Abstract 作者称 LODO 露出 CV 高估,数据集身份与安全方向部分重叠,标准域泛化干预关不上。LODO 用来检查恶意提示词分类器换成没见过的数据集后,还剩多少检测能力。 。
用醉酒语言诱导改编对话模型并评测越狱与隐私泄露
Published Jan 19
Evaluated models LLaMA2-7B , LLaMA3-8B , Mistral-7B and others5 Abstract 醉酒语言诱导在五个 LLM 上抬高越狱与部分隐私错误,作者将其视为攻击向量。作者把醉酒语言当作攻击向量,检验把 LLM 适配成醉酒发短信的模式后,越狱和情境隐私是否变差。动机是人类醉酒时的失范与泄密,以及 LLM 模拟人设的能力。
提出 TS-Guard 与 TS-Flow,在执行前检测并引导 Agent 修正有害工具调用
Published Jan 15
Evaluated models GPT-4o , TS-Guard , Qwen2.5-14B-Instruct Abstract TS-Bench 标步骤级风险,TS-Guard 用多任务 RL 做执行前判定,TS-Flow 用反馈代替直接中止。ToolSafe 研究 LLM 智能体在工具执行前的步骤级安全:基准 TS-Bench、护栏 TS-Guard,以及反馈驱动框架 TS-Flow。