大模型对齐与安全 - arXivDaily 专题

2506.14866 2025-10-30 cs.SE cs.LG 83%

OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents

Thomas Kuntz, Agatha Duzan, Hao Zhao, Francesco Croce, Zico Kolter, Nicolas Flammarion, Maksym Andriushchenko

机构 * EPFL（苏黎世联邦理工学院）； Carnegie Mellon University（卡内基梅隆大学）

专题命中安全评测：safety(title,abstract)；prompt injection(abstract)；分类 cs.LG

Comments NeurIPS 2025 Datasets & Benchmarks Track (Spotlight)

2503.06497 2025-10-30 cs.CV cs.AI 79%

Evaluation of Safety Cognition Capability in Vision-Language Models for Autonomous Driving

Enming Zhang, Peizhe Gong, Xingyuan Dai, Min Huang, Yisheng Lv, Qinghai Miao

机构 * School of Artificial Intelligence, University of Chinese Academy of Sciences（中国科学院大学人工智能学院）； State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所多模态人工智能系统国家重点实验室）

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

2510.19169 2025-10-30 cs.CR cs.CL 70%

OpenGuardrails: A Configurable, Unified, and Scalable Guardrails Platform for Large Language Models

Thomas Wang, Haowen Li

机构 * The Hong Kong Polytechnic University（香港理工大学）

专题命中安全评测：safety(abstract)；prompt injection(abstract)；分类 cs.CL

2510.24898 2025-10-30 eess.SY cs.SY 67%

Delay Tolerant Control for Autonomous Driving Using CDOB

Xincheng Cao, Haochong Chen, Levent Guvenc, Bilin Aksun-Guvenc

专题命中安全评测：alignment(abstract)；safety(abstract)

2510.24811 2025-10-30 cs.CL cs.AI cs.LG 67%

ProofSketch: Efficient Verified Reasoning for Large Language Models

Disha Sheshanarayana, Tanishka Magar

机构 * Manipal University Jaipur（马普尔大学斋普尔）

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

Comments Accepted at NeurIPS 2025, ER Workshop

2510.25771 2025-10-30 cs.CL cs.AI 62%

Gaperon: A Peppered English-French Generative Language Model Suite

Nathan Godey, Wissam Antoun, Rian Touchent, Rachel Bawden, Éric de la Clergerie, Benoît Sagot, Djamé Seddah

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

2510.25571 2025-10-30 cs.LG cs.DS cs.NA math.NA math.SP math.ST stat.TH 57%

Perturbation Bounds for Low-Rank Inverse Approximations under Noise

Phuc Tran, Nisheeth K. Vishnoi

机构 * Yale University（耶鲁大学）

专题命中安全评测：alignment(abstract)；分类 cs.LG

Comments NeurIPS 2025

2510.25413 2025-10-30 cs.CL 57%

Seeing, Signing, and Saying: A Vision-Language Model-Assisted Pipeline for Sign Language Data Acquisition and Curation from Social Media

Shakib Yazdani, Yasser Hamidullah, Cristina España-Bonet, Josef van Genabith

机构 * German Research Center for Artificial Intelligence (DFKI GmbH)（德国人工智能研究中心（DFKI GmbH））； Saarland Informatics Campus（萨尔兰州信息技术校区）； Barcelona Supercomputing Center (BSC-CNS)（巴塞罗那超级计算中心（BSC-CNS））

专题命中安全评测：alignment(abstract)；分类 cs.CL

Comments Accepted by RANLP 2025

2510.25091 2025-10-30 cs.AI 57%

H3M-SSMoEs: Hypergraph-based Multimodal Learning with LLM Reasoning and Style-Structured Mixture of Experts

Peilin Tan, Liang Xie, Churan Zhi, Dian Tu, Chuanqi Shi

机构 * University of California, San Diego（加州大学圣迭戈分校）； Wuhan University of Technology（武汉科技大学）

专题命中安全评测：alignment(abstract)；分类 cs.AI

2510.24951 2025-10-30 cs.LG cs.AR cs.NE 57%

Resource-Efficient and Robust Inference of Deep and Bayesian Neural Networks on Embedded and Analog Computing Platforms

Bernhard Klein

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

Comments Ph.D. dissertation, Heidelberg University, October 2025

2510.24749 2025-10-30 cs.SE cs.AI 57%

Beyond Function-Level Search: Repository-Aware Dual-Encoder Code Retrieval with Adversarial Verification

Aofan Liu, Shiyuan Song, Haoxuan Li, Cehao Yang, Yiyan Qi

机构 * International Digital Economy Academy (IDEA)（国际数字经济学院）； School of Electronic and Computer Engineering, Peking University（电子与计算机工程学院，北京大学）； Shenzhen International Graduate School, Tsinghua University（深圳国际研究生院，清华大学）

专题命中安全评测：alignment(abstract)；分类 cs.AI

Comments Accepted by EMNLP 2025

2509.10266 2025-10-30 cs.CV cs.AI 57%

SignMouth: Leveraging Mouthing Cues for Sign Language Translation by Multimodal Contrastive Fusion

Wenfang Wu, Tingting Yuan, Yupeng Li, Daling Wang, Xiaoming Fu

专题命中安全评测：alignment(abstract)；分类 cs.AI

2510.24999 2025-10-30 cs.CR 50%

SLIP-SEC: Formalizing Secure Protocols for Model IP Protection

Racchit Jain, Satya Lokam, Yehonathan Refael, Adam Hakim, Lev Greenberg, Jay Tenenbaum

专题命中安全评测：safety(abstract)

2506.03089 2025-10-30 cs.CV q-bio.NC 50%

Explicitly Modeling Subcortical Vision with a Neuro-Inspired Front-End Improves CNN Robustness

Lucas Piper, Arlindo L. Oliveira, Tiago Marques

机构 * INESC-ID ； Instituto Superior Técnico（理工学院）； Universidade de Lisboa（里斯本大学）； Breast Cancer Research Program（乳腺癌研究计划）； Champalimaud Foundation（恰帕拉马德基金会）； Faculdade de Medicina de Lisboa（里斯本医学院）

专题命中安全评测：alignment(abstract)

1. 安全评测 14 篇

OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents

Evaluation of Safety Cognition Capability in Vision-Language Models for Autonomous Driving

OpenGuardrails: A Configurable, Unified, and Scalable Guardrails Platform for Large Language Models

Delay Tolerant Control for Autonomous Driving Using CDOB

ProofSketch: Efficient Verified Reasoning for Large Language Models

Gaperon: A Peppered English-French Generative Language Model Suite

Perturbation Bounds for Low-Rank Inverse Approximations under Noise

Seeing, Signing, and Saying: A Vision-Language Model-Assisted Pipeline for Sign Language Data Acquisition and Curation from Social Media

H3M-SSMoEs: Hypergraph-based Multimodal Learning with LLM Reasoning and Style-Structured Mixture of Experts

Resource-Efficient and Robust Inference of Deep and Bayesian Neural Networks on Embedded and Analog Computing Platforms

Beyond Function-Level Search: Repository-Aware Dual-Encoder Code Retrieval with Adversarial Verification

SignMouth: Leveraging Mouthing Cues for Sign Language Translation by Multimodal Contrastive Fusion

SLIP-SEC: Formalizing Secure Protocols for Model IP Protection

Explicitly Modeling Subcortical Vision with a Neuro-Inspired Front-End Improves CNN Robustness