The safety mechanisms of large language models (LLMs) exhibit notable fragility, as even fine-tuning on datasets without harmful content may still undermine their safety capabilities. Meanwhile, existing safety alignment methods predominantly rely on the fine-tuning process, which inadvertently leads to the increased complexity and computational resources required. To address these issues, we introduce LSSF, a novel safety re-alignment framework with \underline{L}ow-Rank \underline{S}afety \underline{S}ubspace \underline{F}usion. Our proposed method exploits the low-rank characteristics of safety information in LLMs by constructing a low-rank projection matrix to extract the principal components of safety vectors. Notably, this projection matrix represents the low-rank safety subspace of the LLMs, which we have observed to remain stable during fine-tuning process and is isolated from the model's general capabilities. These principal components are used to effectively restore safety alignment when combined with fine-tuned LLMs through linear arithmetic. Additionally, to account for the varying encoding densities of safety information across different layers of LLMs, we propose a novel metric called safety singular value entropy. This metric quantifies the encoding density and allows for the dynamic computation of the safety-critical rank for each safety vector. Extensive experiments demonstrate that our proposed post-hoc alignment method can effectively restore the safety alignment of fine-tuned models with minimal impact on their performance in downstream tasks.
Diagnosing and Repairing Unsafe Channels in Vision-Language Models via Causal Discovery and Dual-Modal Safety Subspace Projection
Beijing University of Posts and Telecommunications(北京邮电大学);Huawei Technologies Ltd.(华为技术有限公司);Peking University(北京大学);Chongqing University of Posts and Telecommunications(重庆邮电大学)
A Guardrail for Safety Preservation: When Safety-Sensitive Subspace Meets Harmful-Resistant Null-Space
School of Artificial Intelligence, Jilin University(吉林大学人工智能学院);King Abdullah University of Science and Technology(卡塔尔国王大学科学与技术研究院);University of Oxford(牛津大学)
Safety Subspaces are Not Linearly Distinct: A Fine-Tuning Case Study
Mohamed bin Zayed University of Artificial Intelligence(莫扎德·本·扎耶德人工智能大学);University of California San Diego(加州大学圣地亚哥分校);Massachusetts Institute of Technology(麻省理工学院)
Agent MechSuits: Mechanistic Subspace Safety Steering for Multi-Turn CLI Agents
University of Georgia(佐治亚大学);University of South Florida(南佛罗里达大学);Rutgers University(罗格斯大学);University of Southern California(南加州大学);Johns Hopkins University(约翰霍普金斯大学)
DOG-DPO:Dynamic Optimization in Geometry for Safety Alignment
University of Southern California(南加州大学);Iowa State University(爱荷华州立大学);University of Wisconsin–Madison(威斯康星大学麦迪逊分校);UT Austin(德克萨斯大学奥斯汀分校);Independent Researcher(独立研究员);University of Notre Dame(圣母大学)
Preventing Safety Drift in Large Language Models via Coupled Weight and Activation Constraints
Hunan Normal University(湖南师范大学);The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences(认知与决策智能复杂系统重点实验室,中国科学院自动化研究所);National University of Defense Technology(国防科技大学)
EmoRSS: Mitigating Emotion-Induced Over-Refusal in Large Language Models
Beihang University(北京航空航天大学);University of Science and Technology Beijing(北京科技大学);National University of Singapore(新加坡国立大学);Lanzhou University(兰州大学);Jinan University(暨南大学)
Risk Awareness Injection: Calibrating Vision-Language Models for Safety without Compromising Utility
Shien-Ming Wu School of Intelligent Engineering, South China University of Technology(华南理工大学吴贤铭智能工程学院);Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ)(广东省人工智能与数字经济实验室(深圳));Computer Science and Engineering, The Hong Kong University of Science and Technology(香港科技大学计算机科学与工程系);University of Electronic Science and Technology of China(电子科技大学)
SC-LoRA: Balancing Efficient Fine-tuning and Knowledge Preservation via Subspace-Constrained LoRA
Shanghai Qi Zhi Institute(上海启智研究院);Institute for Interdisciplinary Information Sciences, Tsinghua University(清华大学交叉信息研究院);Institute of Information Engineering, Chinese Academy of Sciences(中国科学院信息工程研究所);Xiongan AI Institute(雄安人工智能研究院)
ProjLens: Unveiling the Role of Projectors in Multimodal Model Safety
University of Science and Technology of China(中国科学技术大学);Beijing University of Aeronautics and Astronautics(北京航空航天大学);Nanyang Technological University(南洋理工大学);Southwest Jiaotong University(西南交通大学);Shanghai Artificial Intelligence Laboratory(上海人工智能实验室)
AlignMerge - Alignment-Preserving Large Language Model Merging via Fisher-Guided Geometric Constraints
AI Institute, University of South Carolina(AI研究院,南卡罗来纳大学);Indian Institute of Technology, Kharagpur(印度理工学院,Khargpur分校);Islamic University of Technology(伊斯兰科技大学);Stanford University, USA(斯坦福大学,美国);Amazon AI, USA(亚马逊AI,美国);HCL(HCL公司);Evalueserve(Evalueserve公司);Apple (USA)(苹果(美国));Google (USA)(谷歌(美国));Pragya Lab, BITS Pilani, Goa(Pragya实验室, BITS Pilani,Goa分校)
Activation Steering Induces Emergent Misalignment: Steering Vector Construction and Comprehensive Evaluation
Nanyang Technological University(南洋理工大学);Sun Yat-sen University(中山大学);University of Science and Technology of China(中国科学技术大学);National University of Singapore(新加坡国立大学)
MSD-LLM: Predicting Ship Detention in Port State Control Inspections with Large Language Model
Institute of High Performance Computing(高性能计算研究所);Agency for Science, Technology and Research(科技研究局);Department of Earth Science and Engineering(地球科学与工程系);Imperial College London(伦敦帝国理工学院);SpaceTimeLab, Department of Civil, Environmental and Geomatic Engineering(空间时间实验室,土木、环境与测绘工程系);University College London(伦敦大学学院);School of Civil and Environmental Engineering(土木与环境工程学院);Nanyang Technological University(南洋理工大学)
MEUV: Achieving Fine-Grained Capability Activation in Large Language Models via Mutually Exclusive Unlock Vectors
Xin Tong School of Information and Cyber Security People’s Public Security University of China(信息与网络安全学院 中国人民公安大学);Zhi Lin School of Safety Science Tsinghua University(安全科学学院 清华大学);Jingya Wang School of Information and Cyber Security People’s Public Security University of China(信息与网络安全学院 中国人民公安大学);Meng Han Intelligent Fusion Research Center Zhejiang University(智能融合研究中心 浙江大学);Bo Jin* The Third Research Institute of the Ministry of Public Security of China(中华人民共和国公安部第三研究所)