Translation in progress Some entries still include text in the original language.
Oct 13 · Tuesday
All papers Found 16 papers Another 201 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.
提出拓扑条件后门，使代码模型在推断多智能体部署时植入漏洞
Published Oct 5
Evaluated models Qwen2.5-7B-Instruct , Qwen2.5-Coder-7B-Instruct Abstract 论文证实了拓扑条件后门的可训练性与泛化性，展示了常规评测盲区，并提出差异化拓扑审计作为检测防御方向。
提出 memetic trojans，借助社交传染在 Agent 网络中传播对抗载荷
Published Oct 1
Evaluated models gpt-oss-120B , deepseek-v4.1-flash , qwen-32B and others5 Abstract memetic trojans 用内生社会传染携带载荷。Memetic trojans 是一类网络介导攻击：对抗载荷搭在智能体本来就会转发和点赞的内容上扩散，转发者不必被攻陷，也不必执行“请继续复制”的指令。
提出只训练或篡改多智能体潜在通信链路来提高有害遵从
Published Sep 30
Evaluated models Llama-3.2-3B-Instruct , Qwen2.5-3B-Instruct , Gemma-3-1B-IT and others7 Abstract 潜在链路本身即可改变冻结智能体系统的安全，奖励既能攻击也能修复。学习得到的潜在通信链路会改变由冻结、对齐智能体组成的系统是否遵从有害请求。 。作者研究三种拓扑：双智能体、三级顺序协作，以及双专家加汇总。
提出 A2A-TIBA，经 A2A 对等任务间接注入并植入常驻回调
Published Sep 30
Evaluated models MiniMax-M3 , deepseek-v4-pro , glm-5.3 and others8 Abstract A2A-TIBA 一次植入后绕过智能体，GDA 用四类结果定位防线，ELA-ITL 把信封标注列为第三层。A2A-TIBA 把间接提示注入和绕过放进 A2A：gray-box 攻击者已能投递任务，诱导目标部署常驻回调程序后，下令和外传不再经过智能体。
提出 BAM 反馈编码，在 LLM 生成文本中实现隐蔽通信
Published Sep 22
Evaluated models Llama-3.1-8B , Qwen-3.5-9B-Base , Mistral-7B-v0.3 and others6 提出 UBA-ORL，以 BD/CM 样本在遗忘后激活离线 RL 后门
Published Sep 19
Evaluated models TD3+BC , BCQ , IQL Abstract UBA-ORL 用竞争的 BD/CM 样本，使离线 RL 后门在删除伪装轨迹后激活。UBA-ORL 研究离线强化学习中由合规轨迹删除激活的后门：合法贡献者同时上传共享触发的后门轨迹和伪装轨迹，训练后触发响应被压低，删除伪装批次后再升高。受害系统是使用静态数据集和轨迹级遗忘接口的离线 RL 服务。对手不能看到训练算法、参数、梯度或具体遗忘算法，但能提交可区分批次，并在部署时向观测注入预定触发。
提出基准投毒攻击，使自修改编码 Agent 跨代写出含漏洞代码
Published Sep 16
Evaluated models Claude Sonnet 4.5 , Qwen3.5-397B , gpt-oss-120b and others5 Abstract 论文报告了恶意基准能污染自修改代码智能体的进化过程并持续输出漏洞，警示系统设计需纳入安全约束。
系统分析智能体支付协议的前置上下文操纵与授权威胁
Published Aug 25 提出多智能体流水线的边界攻击，使对抗内容作为可信输入跨层传播
Published Aug 1
Evaluated models GPT-5-mini , Claude Sonnet 4.5 , Kimi K2.5 Abstract 缺边界校验使对抗内容跨智能体传播，作者称这是架构属性而非模型能力。作者把多智能体 LLM 流水线的对抗问题定义为缺少边界校验，并用生产轨迹和受控实验检查脆弱性来自架构还是骨干模型。中间输出未经验证就传给下游。作者认为，一旦某级接受对抗内容，后续智能体会把它当作可信输入。
提出 MAStrike，用 Shapley 选联盟对多智能体做共谋红队
Published Jun 11
Evaluated models Claude Opus 4.7 , GPT-5.5 , Gemini 3.1 Pro 提出 VPH 与 VTH，只改视频帧劫持世界模型并投毒下游策略
Published Jun 8
Evaluated models Cosmos-Predict 2.5 , Cosmos-Predict 2.5 AC , Dino WM and others4 Abstract 只改视频帧经世界模型投毒。只改看似安全的视频帧，经世界模型之后才改变下游机器人策略。 问题在于世界模型进入数据生成后，数据集检查看到的仍是安全演示，合成轨迹却可以变危险。
建模并利用多智能体网络中子智能体生成时的记忆继承漏洞
Published May 9
Evaluated models MiniMax M2.5 , Llama-4-Maverick-17B-128E-Instruct-FP8 , Qwen3.5-Plus and others5 Abstract 形式化子智能体继承，在 OpenClaw 上演示四类编排层漏洞并提出防御。作者对多智能体网络里的子智能体继承 做形式化分析，并用开源框架上的 PoC 演示编排层如何让单点妥协继续扩散。
Propose a structured JSON backdoor that triggers malicious actions in robot control via a poisoned LoRA
Published Apr 5
Evaluated models Llama-3.1-8B-Instruct , Gemma-2-9B-IT , DeepSeek-R1-Distill-Llama-8B and others4 提出利润驱动红队，用标量盈余搜索对手以压测经济博弈 Agent
Published Mar 22
Evaluated models GPT-OSS-120B , Qwen3-32B , MiniMax-M2.5 and others6 Abstract 利润驱动红队以标量盈余训练对手，六个目标在四博弈中被利用，提示规则可抬高目标盈余。利润驱动红队把可审计终局里的智能体安全写成对手搜索问题：固定目标会不会被一个只追逐自身盈余的自适应玩家利用。
Propose WebWeaver, which compromises only a single agent and recovers the multi-agent communication topology from conversation context.
Published Mar 12
Evaluated models Llama 3.1 8B , Qwen 2.5 7B , Mistral NeMo 12B and others6 Abstract WebWeaver infers topology from context.WebWeaver is an inference attack against the confidentiality of LLM-MAS communication topology: it controls only one arbitrary agent and recovers the full graph from dialogue context, rather than querying agent IDs.
Propose Thought Virus, which propagates preferences along multi-agent messages via subliminal prompts.
Published Feb 23
Evaluated models Qwen2.5-7B-Instruct , Llama-3.1-8B-Instruct Abstract Thought Virus spreads a single-point bias downstream.Thought Virus puts subliminal prompting into multi-agent communication, to see whether a single-point deviation spreads across the network without naming the target concept.
You have reached the end