大模型对齐与安全

Attack the Messages, Not the Agents: A Multi-round Adaptive Stealthy Tampering Framework for LLM-MAS

Bingyu Yan, Ziyi Zhou, Xiaoming Zhang, Chaozhuo Li, Ruilin Zeng, Yirui Qi, Tianbo Wang, Litian Zhang

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Token-Level Precise Attack on RAG: Searching for the Best Alternatives to Mislead Generation

Zizhong Li, Haopeng Zhang, Jiawei Zhang

机构 * University of California, Davis（加州大学戴维斯分校） ； University of Hawaii at Mānoa（夏威夷大学马诺阿分校） 纠错

专题命中 越狱攻击 ：trustworthy(abstract)；分类 cs.CL

Trivial Trojans: How Minimal MCP Servers Enable Cross-Tool Exfiltration of Sensitive Data

Nicola Croce, Tobin South

机构 * Pivotal Research（Pivotal研究机构） ； Stanford University（斯坦福大学） 纠错

专题命中 越狱攻击 ：trustworthy(abstract)；分类 cs.AI

Comments Abstract submitted to the Technical AI Governance Forum 2025 (https://www.techgov.ai/)

ShrinkBox: Backdoor Attack on Object Detection to Disrupt Collision Avoidance in Machine Learning-based Advanced Driver Assistance Systems

Muhammad Zaeem Shahzad, Muhammad Abdullah Hanif, Bassem Ouni, Muhammad Shafique

机构 * eBRAIN Lab, New York University Abu Dhabi (NYUAD), UAE（eBRAIN实验室，纽约大学阿布扎赫尔分校（NYUAD），阿联酋） ； AI and Digital Science Research Center, Technology Innovation Institute (TII), Abu Dhabi, UAE（人工智能与数字科学研究中心，技术创新研究所（TII），阿布扎赫尔，阿联酋） 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.LG

Comments 8 pages, 8 figures, 1 table

Innocence in the Crossfire: Roles of Skip Connections in Jailbreaking Visual Language Models

Palash Nandi, Maithili Joshi, Tanmoy Chakraborty

机构 * Department of Electrical Engineering（电气工程系） ； Indian Institute of Technology Delhi（印度理工学院德里） 纠错

专题命中 越狱攻击 ：jailbreak(abstract)；分类 cs.CL

AI Agent Architecture for Decentralized Trading of Alternative Assets

Ailiya Borjigin, Cong He, Charles CC Lee, Wei Zhou

机构 * Centre for Sustainable Development, University of Newcastle （Australia）, Singapore（可持续发展中心，新南威尔士大学（澳大利亚），新加坡） 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Comments 8 Pages, 1 figure

Bridging AI and Software Security: A Comparative Vulnerability Assessment of LLM Agent Deployment Paradigms

Tarek Gasmi, Ramzi Guesmi, Ines Belhadj, Jihene Bennaceur

专题命中 越狱攻击 ：prompt injection(abstract)；分类 cs.AI

AttentionGuard: Transformer-based Misbehavior Detection for Secure Vehicular Platoons

Hexu Li, Konstantinos Kalogiannis, Ahmed Mohamed Hussain, Panos Papadimitratos

机构 * Networked Systems Security (NSS) Group\ Royal Institute of Technology Stockholm Sweden ； Networked Systems Security (NSS) Group\ Royal Institute of Technology 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Comments Author's version; Accepted for presentation at the ACM Workshop on Wireless Security and Machine Learning (WiseML 2025)

MGC: A Compiler Framework Exploiting Compositional Blindness in Aligned LLMs for Malware Generation

Lu Yan, Zhuo Zhang, Xiangzhe Xu, Shengwei An, Guangyu Shen, Zhou Xuan, Xuan Chen, Xiangyu Zhang

机构 * Purdue University（普渡大学） ； Columbia University（哥伦比亚大学） ； Virginia Tech（弗吉尼亚理工大学） 纠错

专题命中 越狱攻击 ：alignment(abstract)；分类 cs.AI

VSF-Med:A Vulnerability Scoring Framework for Medical Vision-Language Models

Binesh Sadanandan, Vahid Behzadan

机构 * SAIL Lab, University of New Haven, West Haven, CT, USA（SAIL实验室，新罕布什尔大学，西哈文，康涅狄格州，美国） 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Adversarial Threats in Quantum Machine Learning: A Survey of Attacks and Defenses

Archisman Ghosh, Satwik Kundu, Swaroop Ghosh

机构 * Pennsylvania State University（宾夕法尼亚州立大学） 纠错

专题命中 越狱攻击 ：trustworthy(abstract)；分类 cs.LG

Comments 23 pages, 5 figures

Context manipulation attacks : Web agents are susceptible to corrupted memory

Atharv Singh Patlan, Ashwin Hebbar, Pramod Viswanath, Prateek Mittal

机构 * princeton（普林斯顿大学） 纠错

专题命中 越狱攻击 ：prompt injection(abstract)；分类 cs.AI

Comments 10 pages, 6 figures

Stepwise Reasoning Error Disruption Attack of LLMs

Jingyu Peng, Maolin Wang, Xiangyu Zhao, Kai Zhang, Wanyu Wang, Pengyue Jia, Qidong Liu, Ruocheng Guo, Qi Liu

机构 * University of Science and Technology of China（中国科学技术大学） ； City University of Hong Kong（香港城市大学） ； Independent Researcher（独立研究者） 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Investigating Vulnerabilities and Defenses Against Audio-Visual Attacks: A Comprehensive Survey Emphasizing Multimodal Models

Jinming Wen, Xinyi Wu, Shuai Zhao, Yanhao Jia, Yuwen Li

机构 * Jilin University（吉林大学） ； Shanghai Jiao Tong University（上海交通大学） ； Nanyang Technological University（南洋理工大学） ； Northeastern University（东北大学） 纠错

专题命中 越狱攻击 ：jailbreak(abstract)；分类 cs.AI

TokenBreak: Bypassing Text Classification Models Through Token Manipulation

Kasimir Schulz, Kenneth Yeung, Kieran Evans

专题命中 越狱攻击 ：prompt injection(abstract)；分类 cs.LG

Universal Adversarial Attack on Aligned Multimodal LLMs

Temurbek Rahmatullaev, Polina Druzhinina, Nikita Kurdiukov, Matvey Mikhalchuk, Andrey Kuznetsov, Anton Razzhigaev

机构 * AIRI ； MSU（莫斯科国立大学） ； HSE University（俄罗斯高等经济大学） ； Skoltech（斯克里普钦科技大学） 纠错

专题命中 越狱攻击 ：alignment(abstract)；分类 cs.AI

Comments Added benchmarks, baselines, author, appendix

ATAG: AI-Agent Application Threat Assessment with Attack Graphs

Parth Atulbhai Gandhi, Akansha Shukla, David Tayouri, Beni Ifland, Yuval Elovici, Rami Puzis, Asaf Shabtai

机构 * Dept. of Software and Information Systems Engineering（软件与信息系统工程系） ； Ben-Gurion University of the Negev（贝叶尔-加利利大学） 纠错

专题命中 越狱攻击 ：prompt injection(abstract)；分类 cs.AI

Digital Deception: Generative Artificial Intelligence in Social Engineering and Phishing

Marc Schmitt, Ivan Flechais

专题命中 越狱攻击 ：trustworthy(abstract)；分类 cs.CY

Comments Submitted to CHI 2024

Journal ref Artificial Intelligence Review, 2024

Distributed Federated Learning for Vehicular Network Security: Anomaly Detection Benefits and Multi-Domain Attack Threats

Utku Demir, Yalin E. Sagduyu, Tugba Erpek, Hossein Jafari, Sastry Kompella, Mengran Xue

机构 * Nexcepta Inc.（Nexcepta公司） ； RTX BBN Technologies 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Divide and Conquer: A Hybrid Strategy Defeats Multimodal Large Language Models

Yanxu Mao, Peipei Liu, Tiehan Cui, Zhaoteng Yan, Congying Liu, Datao You

机构 * School of Software, Henan University（河南大学软件学院） ； Institute of Information Engineering, Chinese Academy of Sciences（中国科学院信息工程研究所） ； University of Chinese Academy of Sciences（中国科学院大学） 纠错

专题命中 越狱攻击 ：jailbreak(abstract)；分类 cs.CL

What Really Matters in Many-Shot Attacks? An Empirical Study of Long-Context Vulnerabilities in LLMs

Sangyeop Kim, Yohan Lee, Yongwoo Song, Kimin Lee

机构 * Coxwave ； Seoul National University（首尔国立大学） ； Kyung Hee University（庆熙大学） ； KAIST（韩国科学技术院） 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.CL

Comments Accepted by ACL 2025

How Secure Are Large Language Models (LLMs) for Navigation in Urban Environments?

Congcong Wen, Jiazhao Liang, Shuaihang Yuan, Hao Huang, Geeta Chandra Raju Bethala, Yu-Shen Liu, Mengyu Wang, Anthony Tzes, Yi Fang

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Invisible Prompts, Visible Threats: Malicious Font Injection in External Resources for Large Language Models

Junjie Xiong, Changjia Zhu, Shuhang Lin, Chong Zhang, Yongfeng Zhang, Yao Liu, Lingyao Li

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

The Hidden Dangers of Browsing AI Agents

Mykyta Mudryi, Markiyan Chaklosh, Grzegorz Wójcik

机构 * Polish-Japanese Academy of Information Technology（波兰-日本信息科技学院） ； University of the National Education Commission in Kraków（克拉科夫国家教育委员会大学） ； Maria Curie-Sklodowska University in Lublin（利沃夫玛丽亚·克里沃斯卡大学） 纠错

专题命中 越狱攻击 ：prompt injection(abstract)；分类 cs.AI

A Survey of Attacks on Large Language Models

Wenrui Xu, Keshab K. Parhi

机构 * Department of Electrical and Computer Engineering, University of Minnesota（电气与计算机工程系，明尼苏达大学） 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Adversarial Suffix Filtering: a Defense Pipeline for LLMs

David Khachaturov, Robert Mullins

机构 * Department of Computer Science and Technology（计算机科学与技术系） 纠错

专题命中 越狱攻击 ：jailbreak(abstract)；分类 cs.LG

Layered Unlearning for Adversarial Relearning

Timothy Qian, Vinith Suriyakumar, Ashia Wilson, Dylan Hadfield-Menell

机构 * MIT（麻省理工学院） 纠错

专题命中 越狱攻击 ：alignment(abstract)；分类 cs.LG

Comments 37 pages, 8 figures

Endless Jailbreaks with Bijection Learning

Brian R. Y. Huang, Maximilian Li, Leonard Tang

机构 * Haize Labs（哈伊兹实验室） 纠错

专题命中 越狱攻击 ：safety(abstract)；分类 cs.CL

Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks

Han Wang, Gang Wang, Huan Zhang

机构 * University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校） 纠错

专题命中 越狱攻击 ：jailbreak(abstract)；分类 cs.AI

Can We Trust Embodied Agents? Exploring Backdoor Attacks against Embodied LLM-based Decision-Making Systems

Ruochen Jiao, Shaoyuan Xie, Justin Yue, Takami Sato, Lixu Wang, Yixuan Wang, Qi Alfred Chen, Qi Zhu

专题命中 越狱攻击 ：safety(abstract)；分类 cs.AI

Comments Accepted paper at ICLR 2025, 31 pages, including main paper, references, and appendix