LLMs over external data face a per-instruction tension: a single document or tool return often interleaves legitimate directives with adversarial prompt injections, and the model must follow the former while rejecting the latter. Existing defenses suppress or trust all external instructions, collapsing the safety-utility frontier and forcing a choice between over-refusal and over-compliance. We argue the missing capability is \textbf{per-instruction credit attribution} and formalize it as External-Data Instruction Classification, an identify-classify-act decision over each embedded directive. We propose \textbf{EDIC}, a decomposed reward framework for per-instruction compliance. A scene-grounded pipeline injects mixed beneficial and harmful instructions with counterfactual filtering, while a generative judge is initialized by SFT and refined with GRPO to attribute compliance at the instruction level. The judge then produces a reward that separately encourages beneficial compliance and penalizes harmful compliance for RL training. Across two backbones and five benchmarks, EDIC improves both safety and utility over the strongest prior defense, lifting harmful-rejection by up to 7.7\% and Arena-Hard-Auto utility by 12.5\% on Qwen3-4B. The code repository is hosted at https://github.com/yingyingxia666/EDIC.
Compliance vs. Sensibility: On the Reasoning Controllability in Large Language Models
School of Computer Science, University of Sheffield(谢菲尔德大学计算机科学学院);School of EECS, Queen Mary University of London(伦敦女王学院电子工程与计算机科学学院);The Alan Turing Institute(艾伦·图灵研究所)
When Verifiable Counts Depend on Wording: Auditing Wording Robustness in Instruction Following
Marquette University(马凯特大学);University of California, San Diego(加利福尼亚大学圣迭戈分校);Georgia Tech(佐治亚理工学院);Cornell University(康奈尔大学);The University of Texas at Austin(得克萨斯大学奥斯汀分校);Hikvision(海康威视)
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)
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)
Offscript: Automated Auditing of Instruction Adherence in LLMs
University of Washington Information School Seattle Washington USA;University of Washington\ G. Allen School of Computer Science \& Engineering Seattle Washington USA;University of Washington Information School;University of Washington\ G. Allen School of Computer Science \& Engineering
MUFFIN: Curating Multi-Faceted Instructions for Improving Instruction-Following
The Pennsylvania State University(宾夕法尼亚州立大学);The Ohio State University(俄亥俄州立大学);Fudan University(复旦大学);Westlake University(西湖大学);Temple University(天普大学)
InFoBench: Evaluating Instruction Following Ability in Large Language Models
Tencent AI Lab, Seattle(腾讯人工智能实验室(西雅图));University of Central Florida(中佛罗里达大学);University of Georgia(佐治亚大学);Emory University(埃默里大学);Shanghai Jiao Tong University(上海交通大学)
Wait, but Tylenol is Acetaminophen... Investigating and Improving Language Models' Ability to Resist Requests for Misinformation
Mass General Brigham(麻省总医院布莱根医疗系统);Harvard Medical School(哈佛医学院);Brigham and Women’s Hospital(布莱根妇女医院);Dana-Farber Cancer Institute(丹娜-法伯癌症研究所);Boston Children’s Hospital(波士顿儿童医院);Massachusetts Institute of Technology(麻省理工学院);Johns Hopkins University(约翰斯·霍普金斯大学);University of Virginia(弗吉尼亚大学);Maastricht University(马斯特里赫特大学)
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)
The Pushback Paradox: A Two-Probe Diagnostic for Language Model Compliance
Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology(卡尔斯鲁厄理工学院自动化与应用信息学研究所);Duale Hochschule Baden-Württemberg(巴登-符腾堡双元制应用技术大学)
MEIT: Multimodal Electrocardiogram Instruction Tuning on Large Language Models for Report Generation
The Ohio State University(俄亥俄州立大学);Imperial College London(伦敦帝国理工学院);The University of Hong Kong(香港大学);University of North Carolina at Chapel Hill(北卡罗来纳大学教堂山分校)
AISPA: User-Centric System Prompt Auditing for Large Language Model Applications
Stanford University(斯坦福大学);CMU(卡内基梅隆大学);UT Austin(德克萨斯大学奥斯汀分校);University of Toronto(多伦多大学);UCSB(加利福尼亚大学圣巴巴拉分校);WashU(华盛顿大学);OSU(俄亥俄州立大学);UCSC(加利福尼亚大学圣克鲁兹分校);Northwestern University(西北大学);UIUC(伊利诺伊大学厄巴纳-香槟分校);KAUST(阿卜杜拉国王科技大学);MIT(麻省理工学院);University of Oxford(牛津大学);Institute for Decentralized AI(去中心化人工智能研究所)
Reference-Grounded Data Curation for Instruction-Following Thai-English Machine Translation
iApp Technology(iApp科技公司);Intelligent Informatics and Service Innovation Research Center(智能信息学与服务创新研究中心);Artificial Intelligence Entrepreneur Association of Thailand (AIEAT)(泰国人工智能企业家协会);Sirindhorn International Institute of Technology, Thammasat University(诗琳通国际理工学院,法政大学)