It's a TRAP! Task-Redirecting Agent Persuasion Benchmark for Web Agents
这是一个陷阱!面向网络代理的任务重定向说服基准
Karolina Korgul, Yushi Yang, Arkadiusz Drohomirecki, Piotr Błaszczyk, Will Howard, Lukas Aichberger, Chris Russell, Philip H. S. Torr, Adam Mahdi, Adel Bibi
The Surface You Test Is Not the Surface That Breaks
测试的表面并非断裂的表面
Shifat E Arman, Syed Nazmus Sakib, Nafiul Haque, Shahrear Bin Amin
机构
*
Department of Robotics and Mechatronics Engineering, University of Dhaka(达卡大学机器人与机电工程系)
;
Department of Computer Science and Engineering, University of Dhaka(达卡大学计算机科学与工程系)
专题命中
提示注入
:prompt injection(abstract);分类 cs.AI
AI总结
本文发现工具增强的LLM代理对提示注入的脆弱性依赖于攻击表面(工具输出 vs 工具描述),提出自适应攻击率并强调评估需报告每个表面的脆弱性。
Comments8 Figures, 8 Tables, Under Review at EMNLP
Hidden in Memory: Sleeper Memory Poisoning in LLM Agents
记忆中的隐患:LLM代理中的潜伏记忆污染
Sidharth Pulipaka, Stanislau Hlebik, Leonidas Raghav, Sahar Abdelnabi, Vyas Raina, Ivaxi Sheth, Mario Fritz
机构
*
SPAR
;
ELLIS Institute Tübingen(图宾根ELLIS研究所)
;
MPI for Intelligent Systems(智能系统马克斯·普朗克研究所)
;
Tübingen AI Center(图宾根人工智能中心)
;
APTA
;
CISPA Helmholtz Center for Information Security(信息安全海德堡中心)
机构
*
Gaoling School of Artificial Intelligence, Renmin University of China(中国人民大学人工智能学院 Gallagher 学院)
;
King Abdullah University of Science and Technology(国王 Abdullah 科学技术大学)
;
Dongbei University of Finance and Economics(东北财经大学)
;
University of Science and Technology of China(中国科学技术大学)
AgentShield: Deception-based Compromise Detection for Tool-using LLM Agents
AgentShield:基于欺骗的工具使用LLM代理 compromise 检测
Yassin H. Rassul, Tarik A. Rashid
机构
*
Computer Science and Engineering Department, School of Science and Engineering, University of Kurdistan Hewl\^{e}r, Erbil, Iraq(科罗曼嫩大学科学与工程学院计算机科学与工程系)
;
Engineering Department, School of Science(科学学院工程系)
;
Engineering, University of Kurdistan Hewl\ e r, Erbil, Iraq(科罗曼嫩大学工程学院)
;
Innovation Centre, University of Kurdistan Hewl\ e r, Erbil, Iraq(科罗曼嫩大学创新中心)