Large language model agents receive instructions from many sources-system messages, user prompts, tool outputs, other agents, and more-each carrying different levels of trust and authority. When these instructions conflict, agents must reliably follow the highest-privilege instruction to remain safe and effective. The dominant paradigm, instruction hierarchy (IH), assumes a fixed, small set of privilege levels (typically fewer than five) defined by rigid role labels (e.g., system > user). This is inadequate for real-world agentic settings, where conflicts can arise across far more sources and contexts. In this work, we propose Many-Tier Instruction Hierarchy (ManyIH), a paradigm for resolving instruction conflicts among instructions with arbitrarily many privilege levels. We introduce ManyIH-Bench, the first benchmark for ManyIH. ManyIH-Bench requires models to navigate up to 12 levels of conflicting instructions with varying privileges, comprising 853 agentic tasks (427 coding and 426 instruction-following). ManyIH-Bench composes constraints developed by LLMs and verified by humans to create realistic and difficult test cases spanning 46 real-world agents. Our experiments show that even the current frontier models perform poorly (~40% accuracy) when instruction conflict scales. This work underscores the urgent need for methods that explicitly target fine-grained, scalable instruction conflict resolution in agentic settings.
Reasoning Up the Instruction Ladder for Controllable Language Models
Department of Computer Science and Engineering, The Ohio State University（俄亥俄州立大学计算机科学与工程系）；Microsoft Research（微软研究院）；Allen Institute for AI（艾伦人工智能研究所）
HIPO: Instruction Hierarchy via Constrained Reinforcement Learning
School of ECEE Arizona State University（亚利桑那州立大学电子与计算机工程学院）；Department of ECE The Ohio State University（俄亥俄州立大学电子工程系）；Computer Science Department University of Houston（休斯顿大学计算机科学系）；College of Engineering & Applied Science CU Boulder（科罗拉多大学博尔德分校工程与应用科学学院）；Dept. of Electrical Engineering & Computer Science United States Military Academy（美国军事学院电子工程与计算机科学系）
Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents
The Pennsylvania State University（宾夕法尼亚州立大学）；UC Santa Cruz（加州大学圣克鲁兹分校）；Amazon（亚马逊）；Emory University（埃默里大学）；UIUC（伊利诺伊大学香槟分校）；Northeastern University（东北大学）
Financial Services Innovation Lab, Georgia Institute of Technology（佐治亚理工学院金融服务创新实验室）；College of Business, Georgia Institute of Technology（佐治亚理工学院商学院）；College of Computing, Georgia Institute of Technology（佐治亚理工学院计算学院）
Evaluating and Aligning CodeLLMs on Human Preference
Alibaba Group（阿里巴巴集团）；Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences（中国科学院深圳先进技术研究院）；University of Chinese Academy of Sciences（中国科学院大学）；Shanghai Jiao Tong University（上海交通大学）
Diagnose, Localize, Align: A Full-Stack Framework for Reliable LLM Multi-Agent Systems under Instruction Conflicts
University of California, Los Angeles（加州大学洛杉矶分校）；Sea AI Lab（Sea AI 实验室）；Stanford University（斯坦福大学）；University of Oxford（牛津大学）；Yale University（耶鲁大学）；NTU（南洋理工大学）；NUS（新加坡国立大学）；Boston University（波士顿大学）
From Retrieval to Reasoning: A Framework for Cyber Threat Intelligence NER with Explicit and Adaptive Instructions
Sichuan University（四川大学）；Renmin University of China（中国人民大学）；Wuhan University（武汉大学）；Engineering Research Center of Next-Generation Intelligent Search and Recommendation, Ministry of Education, China（教育部下一代智能搜索与推荐工程研究中心）
Learning Invariable Semantical Representation from Language for Extensible Policy Generalization
Tsinghua University（清华大学）；Beijing Innovation Center for Future Chip（北京未来芯片创新中心）；LSBDPA Beijing Key Laboratory（北京市LSBDPA重点实验室）；North China University of Technology（北方工业大学）
