LLM agents increasingly run inside execution harnesses that dispatch tools, allocate resources, and route messages between specialized components. However, a harness can return a correct, benign answer over a trajectory that accesses unauthorized resources or leaks context to the wrong agent. Output-level evaluation cannot see these failures, yet most safety benchmarks score only final outputs or terminal states, even though many violations occur mid-trajectory rather than at termination. The central question is whether the harness respects user intent, permission boundaries, and information-flow constraints throughout execution. To address this gap, we propose HarnessAudit, a framework that audits full execution trajectories across boundary compliance, execution fidelity, and system stability, with a focus on multi-agent harnesses where these risks are most pronounced. We further introduce HarnessAudit-Bench, a benchmark of 210 tasks across eight real-world domains, instantiated in both single-agent and multi-agent configurations with embedded safety constraints. Evaluating ten harness configurations across frontier models and three multi-agent frameworks, we find that: (i) task completion is misaligned with safe execution, and violations accumulate with trajectory length; (ii) safety risks vary across domains, task types, and agent roles; (iii) most violations concentrate in resource access and inter-agent information transfer; and (iv) multi-agent collaboration expands the safety risk surface, while harness design sets the upper bound of safe deployment.
FinHarness: An Inline Lifecycle Safety Harness for Finance LLM Agents
Peking University(北京大学);Nanyang Technological University(南洋理工大学);Tsinghua University(清华大学);University of Science and Technology of China(中国科学技术大学);University of Chinese Academy of Sciences(中国科学院大学);Soochow University(苏州大学);Beijing University of Posts and Telecommunications(北京邮电大学);University of Leeds(利兹大学);Fullive Innovation (Beijing) AI Technology Co., Ltd.(全维创新(北京)人工智能科技有限公司);Beihang University(北京航空航天大学);University of Illinois Chicago(伊利诺伊大学芝加哥分校)
Inform, Coach, Relate, Listen: Auditing LLM Caregiving Support Roles
University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);University of Massachusetts Amherst(马萨诸塞大学阿默斯特分校);OSF HealthCare(OSF医疗集团);Indiana University Indianapolis(印第安纳大学印第安纳波利斯分校)
Harnessing Task Overload for Scalable Jailbreak Attacks on Large Language Models
School of Future Technology, University of Chinese Academy of Sciences(中国科学院大学未来技术学院);Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所);Beijing Institute of AI Safety and Governance(北京人工智能安全与治理研究院);Center for Long-term Artificial Intelligence(长期人工智能中心);School of Artificial Intelligence, University of Chinese Academy of Sciences(中国科学院大学人工智能学院)
From Automation to Collaboration: Human-in-the-Loop Methods for Safe and Trustworthy NLP
Department of Computer Science and Engineering, BRAC University(布拉克大学计算机科学与工程系);Department of Electrical and Electronic Engineering, Rajshahi University of Engineering and Technology(拉贾沙希工程与技术大学电子与电气工程系)
Colosseum: Auditing Collusion in Cooperative Multi-Agent Systems
University of Massachusetts Amherst(马萨诸塞大学阿姆赫斯特分校);University of Virginia(弗吉尼亚大学);ELLIS Institute Tübingen(图宾根ELLIS研究所);MPI for Intelligent Systems, Tübingen(图宾根智能系统研究所);AI Center(人工智能中心)
DoubtProbe: Black-Box Jailbreak Defense via Structural Verification and Semantic Auditing
East China Normal University(东华大学);The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州));Institute of Information Engineering, Chinese Academy of Sciences(中国科学院信息工程研究所)
Auditing medical multi-agent AI reveals risks of false consensus
National Engineering Research Center for Software Engineering, Peking University(北京大学软件工程国家工程研究中心);School of Computing and Data Science, The University of Hong Kong(香港大学计算机与数据科学学院);Department of Nephrology, Peking University Third Hospital(北京大学第三医院肾内科);Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education), Department of Lymphoma, Peking University Cancer Hospital & Institute(教育部癌症发生与转化研究重点实验室、北京大学肿瘤医院淋巴瘤科);Department of Automation, Tsinghua University(清华大学自动化系);Centre for Medical Informatics, The University of Edinburgh(爱丁堡大学医学信息学中心);Health Data Research UK(英国健康数据研究机构);Lee Kong Chian School of Medicine, Nanyang Technological University(南洋理工大学李科贤医学院);State Key Laboratory for Novel Software Technology, School of Computer Science, Nanjing University(南京大学新型软件技术国家重点实验室、计算机科学学院)