大模型对齐与安全

The Secret Agenda: LLMs Strategically Lie and Our Current Safety Tools Are Blind

Caleb DeLeeuw, Gaurav Chawla, Aniket Sharma, Vanessa Dietze

机构 * Independent Researcher（独立研究者） 纠错

专题命中 安全评测 ：safety(title)；分类 cs.AI、cs.CY、cs.LG

Comments 9 pages plus citations and appendix, 7 figures

Can Federated Learning Safeguard Private Data in LLM Training? Vulnerabilities, Attacks, and Defense Evaluation

Wenkai Guo, Xuefeng Liu, Haolin Wang, Jianwei Niu, Shaojie Tang, Jing Yuan

机构 * State Key Laboratory of Virtual Reality Technology and Systems, School of Computer Science and Engineering, Beihang University, Beijing, China（虚拟现实技术与系统国家重点实验室，计算机科学与工程学院，北京航空航天大学） ； Hangzhou Innovation Institute of Beihang University, Zhejiang Key Laboratory of Industrial Big Data and Robot Intelligent Systems, Hangzhou, China（北京航空航天大学杭州创新研究院，浙江省工业大数据与机器人智能系统重点实验室） ； Center for AI Business Innovation, Department of Management Science and Systems, University at Buffalo, Buffalo, New York, USA（人工智能商业创新中心，管理科学与系统系，布法罗大学） ； University of North Texas, Denton, Texas, USA（德克萨斯大学达文波特分校） ； Zhongguancun Laboratory, Beijing, China（中关村实验室） 纠错

专题命中 安全评测 ：alignment(abstract)；safety(abstract)；分类 cs.CL、cs.LG

Comments 28 pages, 32 figures, accepted to the Findings of EMNLP 2025

CORE: Full-Path Evaluation of LLM Agents Beyond Final State

Panagiotis Michelakis, Yiannis Hadjiyiannis, Dimitrios Stamoulis

机构 * Synkrasis Labs（Synkrasis实验室） ； Harbin Institute of Technology（哈尔滨工业大学） 纠错

专题命中 安全评测 ：alignment(abstract)；safety(abstract)；分类 cs.AI

Comments Accepted: LAW 2025 Workshop NeurIPS 2025

DisCoCLIP: A Distributional Compositional Tensor Network Encoder for Vision-Language Understanding

Kin Ian Lo, Hala Hawashin, Mina Abbaszadeh, Tilen Limback-Stokin, Hadi Wazni, Mehrnoosh Sadrzadeh

机构 * University College London（伦敦大学学院） 纠错

专题命中 安全评测 ：alignment(abstract)；分类 cs.CL、cs.AI

Adaptive Approach to Enhance Machine Learning Scheduling Algorithms During Runtime Using Reinforcement Learning in Metascheduling Applications

Samer Alshaer, Ala Khalifeh, Roman Obermaisser

专题命中 安全评测 ：safety(abstract)；分类 cs.AI、cs.LG

Comments 18 pages, 21 figures

Beyond Global Emotion: Fine-Grained Emotional Speech Synthesis with Dynamic Word-Level Modulation

Sirui Wang, Andong Chen, Tiejun Zhao

机构 * Harbin Institute of Technology（哈尔滨工业大学） 纠错

专题命中 安全评测 ：trustworthy(abstract)；分类 cs.CL、cs.AI

The Validation Gap: A Mechanistic Analysis of How Language Models Compute Arithmetic but Fail to Validate It

Leonardo Bertolazzi, Philipp Mondorf, Barbara Plank, Raffaella Bernardi

机构 * DISI, University of Trento（特伦托大学DISI中心） ； MaiNLP, Center for Information and Language Processing, LMU Munich（慕尼黑大学信息与语言处理中心） ； Munich Center for Machine Learning (MCML), Munich, Germany（慕尼黑机器学习中心） ； Free University of Bozen-Bolzano, Italy（博兹纳自由大学） 纠错

专题命中 安全评测 ：alignment(abstract)；分类 cs.CL、cs.AI

Comments EMNLP 2025 Main, 38 pages, 33 figures

SD3.5-Flash: Distribution-Guided Distillation of Generative Flows

Hmrishav Bandyopadhyay, Rahim Entezari, Jim Scott, Reshinth Adithyan, Yi-Zhe Song, Varun Jampani

机构 * Stability AI ； SketchX, University of Surrey（SketchX，大学） 纠错

专题命中 安全评测 ：alignment(abstract)；分类 cs.AI

Comments Project Page: https://hmrishavbandy.github.io/sd35flash/

SAGE: A Realistic Benchmark for Semantic Understanding

Samarth Goel, Reagan J. Lee, Kannan Ramchandran

机构 * University of California, Berkeley（加州大学伯克利分校） 纠错

专题命中 安全评测 ：alignment(abstract)；分类 cs.AI

Comments 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: Evaluating the Evolving LLM Lifecycle: Benchmarks, Emergent Abilities, and Scaling

CLaw: Benchmarking Chinese Legal Knowledge in Large Language Models - A Fine-grained Corpus and Reasoning Analysis

Xinzhe Xu, Liang Zhao, Hongshen Xu, Chen Chen

机构 * Peking University（北京大学） ； LLM-Core 纠错

专题命中 安全评测 ：trustworthy(abstract)；分类 cs.CL

A Taxonomy of Data Risks in AI and Quantum Computing (QAI) - A Systematic Review

Grace Billiris, Asif Gill, Madhushi Bandara

专题命中 安全评测 ：trustworthy(abstract)；分类 cs.AI

Comments 11 pages, 2 figures, 2 tables

Investigating Traffic Accident Detection Using Multimodal Large Language Models

Ilhan Skender, Kailin Tong, Selim Solmaz, Daniel Watzenig

机构 * Embedded Systems Group (Dept.-E)（嵌入式系统组） ； Virtual Vehicle Research GmbH（虚拟车辆研究公司） ； Control Systems Group (Dept.-E)（控制系统组） ； Institute of Visual Computing（视觉计算研究所） ； Graz University of Technology（格拉茨技术大学） 纠错

专题命中 安全评测 ：safety(abstract)

Comments Accepted for presentation at the 2025 IEEE International Automated Vehicle Validation Conference (IAVVC 2025). Final version to appear in IEEE Xplore