Large language models (LLMs) and multimodal LLMs are typically safety-aligned before release to prevent harmful content generation. However, recent studies show that safety behaviors are concentrated in a small subset of parameters, making alignment brittle and easily bypassed through neuron-level attacks. Moreover, most existing alignment methods operate at the behavioral level, offering limited control over the model's internal safety mechanisms. In this work, we propose SafeNeuron, a neuron-level safety alignment framework that improves robustness by redistributing safety representations across the network. SafeNeuron first identifies safety-related neurons, then freezes these neurons during preference optimization to prevent reliance on sparse safety pathways and force the model to construct redundant safety representations. Extensive experiments across models and modalities demonstrate that SafeNeuron significantly improves robustness against neuron pruning attacks, reduces the risk of open-source models being repurposed as red-team generators, and preserves general capabilities. Furthermore, our layer-wise analysis reveals that safety behaviors are governed by stable and shared internal representations. Overall, SafeNeuron provides an interpretable and robust perspective for model alignment.
Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons
Department of Computer Science and Technology, BNRist(计算机科学与技术系,BNRist);Shenzhen International Graduate School(深圳国际研究生院);KIRC, Institute for Artificial Intelligence, Tsinghua University(人工智能研究院,清华大学)
Light Alignment Improves LLM Safety via Model Self-Reflection with a Single Neuron
Beijing Institute of AI Safety and Governance(北京人工智能安全与治理研究院);Beijing Key Laboratory of Safe AI and Superalignment(北京安全人工智能与超对齐重点实验室);Ant Group Co., Ltd.(蚂蚁集团有限公司);BrainCog Lab., CASIA(CASIA脑认知实验室);Zhongguancun Academy(中关村学院)
SafeSeek: Universal Attribution of Safety Circuits in Language Models
University of Science;United Arab Emirates University;Nanyang Technological University;Zayed University;Intelligent Science \& Technology Academy of CASIC
Under the Hood of Neural Networks: Characterizing Learned Representations by Functional Neuron Populations and Network Ablations
Institute of Technologies and Management of Digital Transformation(数字化转型技术与管理研究所);Institute of Information Management in Mechanical Engineering(机械工程信息管理研究所)
Multi-Level Safety Continual Projection for Fine-Tuned Large Language Models without Retraining
Beijing Key Laboratory of Safe AI and Superalignment(北京安全人工智能与超对齐重点实验室);Beijing Institute of AI Safety and Governance(北京人工智能安全与治理研究院);Brain-inspired Cognitive AI Lab, Institute of Automation, Chinese Academy of Sciences(脑启发认知人工智能实验室,中国科学院自动化研究所);University of Chinese Academy of Sciences(中国科学院大学);Long-term AI(长期人工智能)
Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models
Technical University of Munich(慕尼黑技术大学);Helmholtz Munich(亥姆霍兹慕尼黑);Munich Center for Machine Learning(慕尼黑机器学习中心);Munich Data Science Institute(慕尼黑数据科学研究所);University of Tübingen(图宾根大学);University of Copenhagen(哥本哈根大学)
Representation Engineering: A Top-Down Approach to AI Transparency
Center for AI Safety(人工智能安全中心);Carnegie Mellon University(卡内基梅隆大学);Stanford University(斯坦福大学);Cornell University(康奈尔大学);University of Maryland(马里兰大学);University of Pennsylvania(宾夕法尼亚大学);UC Berkeley(加州大学伯克利分校);University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);EleutherAI