Buchumschlag

A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment

Fuente: arXiv

Gespeichert in:

Bibliographische Detailangaben
Hauptverfasser:	Wang, Kun, Zhang, Guibin, Zhou, Zhenhong, Wu, Jiahao, Yu, Miao, Zhao, Shiqian, Yin, Chenlong, Fu, Jinhu, Yan, Yibo, Luo, Hanjun, Lin, Liang, Xu, Zhihao, Lu, Haolang, Cao, Xinye, Zhou, Xinyun, Jin, Weifei, Meng, Fanci, Xu, Shicheng, Mao, Junyuan, Wang, Yu, Wu, Hao, Wang, Minghe, Zhang, Fan, Fang, Junfeng, Qu, Wenjie, Liu, Yue, Liu, Chengwei, Zhang, Yifan, Li, Qiankun, Guo, Chongye, Qin, Yalan, Fan, Zhaoxin, Wang, Kai, Ding, Yi, Hong, Donghai, Ji, Jiaming, Lai, Yingxin, Yu, Zitong, Li, Xinfeng, Jiang, Yifan, Li, Yanhui, Deng, Xinyu, Wu, Junlin, Wang, Dongxia, Huang, Yihao, Guo, Yufei, Huang, Jen-tse, Wang, Qiufeng, Jin, Xiaolong, Wang, Wenxuan, Liu, Dongrui, Yue, Yanwei, Huang, Wenke, Wan, Guancheng, Chang, Heng, Li, Tianlin, Yu, Yi, Li, Chenghao, Li, Jiawei, Bai, Lei, Zhang, Jie, Guo, Qing, Wang, Jingyi, Chen, Tianlong, Zhou, Joey Tianyi, Jia, Xiaojun, Sun, Weisong, Wu, Cong, Chen, Jing, Hu, Xuming, Li, Yiming, Wang, Xiao, Zhang, Ningyu, Tuan, Luu Anh, Xu, Guowen, Zhang, Jiaheng, Zhang, Tianwei, Ma, Xingjun, Gu, Jindong, Pang, Liang, Wang, Xiang, An, Bo, Sun, Jun, Bansal, Mohit, Pan, Shirui, Lyu, Lingjuan, Elovici, Yuval, Kailkhura, Bhavya, Yang, Yaodong, Li, Hongwei, Xu, Wenyuan, Sun, Yizhou, Wang, Wei, Li, Qing, Tang, Ke, Jiang, Yu-Gang, Juefei-Xu, Felix, Xiong, Hui, Wang, Xiaofeng, Tao, Dacheng, Yu, Philip S., Wen, Qingsong, Liu, Yang
Format:	Preprint
Veröffentlicht:	2025
Schlagworte:	Cryptography and Security Artificial Intelligence Computation and Language Machine Learning
Online-Zugang:	Acceder al recurso
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

_version_	1866908398475280384
author	Wang, Kun Zhang, Guibin Zhou, Zhenhong Wu, Jiahao Yu, Miao Zhao, Shiqian Yin, Chenlong Fu, Jinhu Yan, Yibo Luo, Hanjun Lin, Liang Xu, Zhihao Lu, Haolang Cao, Xinye Zhou, Xinyun Jin, Weifei Meng, Fanci Xu, Shicheng Mao, Junyuan Wang, Yu Wu, Hao Wang, Minghe Zhang, Fan Fang, Junfeng Qu, Wenjie Liu, Yue Liu, Chengwei Zhang, Yifan Li, Qiankun Guo, Chongye Qin, Yalan Fan, Zhaoxin Wang, Kai Ding, Yi Hong, Donghai Ji, Jiaming Lai, Yingxin Yu, Zitong Li, Xinfeng Jiang, Yifan Li, Yanhui Deng, Xinyu Wu, Junlin Wang, Dongxia Huang, Yihao Guo, Yufei Huang, Jen-tse Wang, Qiufeng Jin, Xiaolong Wang, Wenxuan Liu, Dongrui Yue, Yanwei Huang, Wenke Wan, Guancheng Chang, Heng Li, Tianlin Yu, Yi Li, Chenghao Li, Jiawei Bai, Lei Zhang, Jie Guo, Qing Wang, Jingyi Chen, Tianlong Zhou, Joey Tianyi Jia, Xiaojun Sun, Weisong Wu, Cong Chen, Jing Hu, Xuming Li, Yiming Wang, Xiao Zhang, Ningyu Tuan, Luu Anh Xu, Guowen Zhang, Jiaheng Zhang, Tianwei Ma, Xingjun Gu, Jindong Pang, Liang Wang, Xiang An, Bo Sun, Jun Bansal, Mohit Pan, Shirui Lyu, Lingjuan Elovici, Yuval Kailkhura, Bhavya Yang, Yaodong Li, Hongwei Xu, Wenyuan Sun, Yizhou Wang, Wei Li, Qing Tang, Ke Jiang, Yu-Gang Juefei-Xu, Felix Xiong, Hui Wang, Xiaofeng Tao, Dacheng Yu, Philip S. Wen, Qingsong Liu, Yang
author_facet	Wang, Kun Zhang, Guibin Zhou, Zhenhong Wu, Jiahao Yu, Miao Zhao, Shiqian Yin, Chenlong Fu, Jinhu Yan, Yibo Luo, Hanjun Lin, Liang Xu, Zhihao Lu, Haolang Cao, Xinye Zhou, Xinyun Jin, Weifei Meng, Fanci Xu, Shicheng Mao, Junyuan Wang, Yu Wu, Hao Wang, Minghe Zhang, Fan Fang, Junfeng Qu, Wenjie Liu, Yue Liu, Chengwei Zhang, Yifan Li, Qiankun Guo, Chongye Qin, Yalan Fan, Zhaoxin Wang, Kai Ding, Yi Hong, Donghai Ji, Jiaming Lai, Yingxin Yu, Zitong Li, Xinfeng Jiang, Yifan Li, Yanhui Deng, Xinyu Wu, Junlin Wang, Dongxia Huang, Yihao Guo, Yufei Huang, Jen-tse Wang, Qiufeng Jin, Xiaolong Wang, Wenxuan Liu, Dongrui Yue, Yanwei Huang, Wenke Wan, Guancheng Chang, Heng Li, Tianlin Yu, Yi Li, Chenghao Li, Jiawei Bai, Lei Zhang, Jie Guo, Qing Wang, Jingyi Chen, Tianlong Zhou, Joey Tianyi Jia, Xiaojun Sun, Weisong Wu, Cong Chen, Jing Hu, Xuming Li, Yiming Wang, Xiao Zhang, Ningyu Tuan, Luu Anh Xu, Guowen Zhang, Jiaheng Zhang, Tianwei Ma, Xingjun Gu, Jindong Pang, Liang Wang, Xiang An, Bo Sun, Jun Bansal, Mohit Pan, Shirui Lyu, Lingjuan Elovici, Yuval Kailkhura, Bhavya Yang, Yaodong Li, Hongwei Xu, Wenyuan Sun, Yizhou Wang, Wei Li, Qing Tang, Ke Jiang, Yu-Gang Juefei-Xu, Felix Xiong, Hui Wang, Xiaofeng Tao, Dacheng Yu, Philip S. Wen, Qingsong Liu, Yang
contents	The remarkable success of Large Language Models (LLMs) has illuminated a promising pathway toward achieving Artificial General Intelligence for both academic and industrial communities, owing to their unprecedented performance across various applications. As LLMs continue to gain prominence in both research and commercial domains, their security and safety implications have become a growing concern, not only for researchers and corporations but also for every nation. Currently, existing surveys on LLM safety primarily focus on specific stages of the LLM lifecycle, e.g., deployment phase or fine-tuning phase, lacking a comprehensive understanding of the entire "lifechain" of LLMs. To address this gap, this paper introduces, for the first time, the concept of "full-stack" safety to systematically consider safety issues throughout the entire process of LLM training, deployment, and eventual commercialization. Compared to the off-the-shelf LLM safety surveys, our work demonstrates several distinctive advantages: (I) Comprehensive Perspective. We define the complete LLM lifecycle as encompassing data preparation, pre-training, post-training, deployment and final commercialization. To our knowledge, this represents the first safety survey to encompass the entire lifecycle of LLMs. (II) Extensive Literature Support. Our research is grounded in an exhaustive review of over 800+ papers, ensuring comprehensive coverage and systematic organization of security issues within a more holistic understanding. (III) Unique Insights. Through systematic literature analysis, we have developed reliable roadmaps and perspectives for each chapter. Our work identifies promising research directions, including safety in data generation, alignment techniques, model editing, and LLM-based agent systems. These insights provide valuable guidance for researchers pursuing future work in this field.
format	Preprint
id	arxiv_https___arxiv_org_abs_2504_15585
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment Wang, Kun Zhang, Guibin Zhou, Zhenhong Wu, Jiahao Yu, Miao Zhao, Shiqian Yin, Chenlong Fu, Jinhu Yan, Yibo Luo, Hanjun Lin, Liang Xu, Zhihao Lu, Haolang Cao, Xinye Zhou, Xinyun Jin, Weifei Meng, Fanci Xu, Shicheng Mao, Junyuan Wang, Yu Wu, Hao Wang, Minghe Zhang, Fan Fang, Junfeng Qu, Wenjie Liu, Yue Liu, Chengwei Zhang, Yifan Li, Qiankun Guo, Chongye Qin, Yalan Fan, Zhaoxin Wang, Kai Ding, Yi Hong, Donghai Ji, Jiaming Lai, Yingxin Yu, Zitong Li, Xinfeng Jiang, Yifan Li, Yanhui Deng, Xinyu Wu, Junlin Wang, Dongxia Huang, Yihao Guo, Yufei Huang, Jen-tse Wang, Qiufeng Jin, Xiaolong Wang, Wenxuan Liu, Dongrui Yue, Yanwei Huang, Wenke Wan, Guancheng Chang, Heng Li, Tianlin Yu, Yi Li, Chenghao Li, Jiawei Bai, Lei Zhang, Jie Guo, Qing Wang, Jingyi Chen, Tianlong Zhou, Joey Tianyi Jia, Xiaojun Sun, Weisong Wu, Cong Chen, Jing Hu, Xuming Li, Yiming Wang, Xiao Zhang, Ningyu Tuan, Luu Anh Xu, Guowen Zhang, Jiaheng Zhang, Tianwei Ma, Xingjun Gu, Jindong Pang, Liang Wang, Xiang An, Bo Sun, Jun Bansal, Mohit Pan, Shirui Lyu, Lingjuan Elovici, Yuval Kailkhura, Bhavya Yang, Yaodong Li, Hongwei Xu, Wenyuan Sun, Yizhou Wang, Wei Li, Qing Tang, Ke Jiang, Yu-Gang Juefei-Xu, Felix Xiong, Hui Wang, Xiaofeng Tao, Dacheng Yu, Philip S. Wen, Qingsong Liu, Yang Cryptography and Security Artificial Intelligence Computation and Language Machine Learning The remarkable success of Large Language Models (LLMs) has illuminated a promising pathway toward achieving Artificial General Intelligence for both academic and industrial communities, owing to their unprecedented performance across various applications. As LLMs continue to gain prominence in both research and commercial domains, their security and safety implications have become a growing concern, not only for researchers and corporations but also for every nation. Currently, existing surveys on LLM safety primarily focus on specific stages of the LLM lifecycle, e.g., deployment phase or fine-tuning phase, lacking a comprehensive understanding of the entire "lifechain" of LLMs. To address this gap, this paper introduces, for the first time, the concept of "full-stack" safety to systematically consider safety issues throughout the entire process of LLM training, deployment, and eventual commercialization. Compared to the off-the-shelf LLM safety surveys, our work demonstrates several distinctive advantages: (I) Comprehensive Perspective. We define the complete LLM lifecycle as encompassing data preparation, pre-training, post-training, deployment and final commercialization. To our knowledge, this represents the first safety survey to encompass the entire lifecycle of LLMs. (II) Extensive Literature Support. Our research is grounded in an exhaustive review of over 800+ papers, ensuring comprehensive coverage and systematic organization of security issues within a more holistic understanding. (III) Unique Insights. Through systematic literature analysis, we have developed reliable roadmaps and perspectives for each chapter. Our work identifies promising research directions, including safety in data generation, alignment techniques, model editing, and LLM-based agent systems. These insights provide valuable guidance for researchers pursuing future work in this field.
title	A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment
topic	Cryptography and Security Artificial Intelligence Computation and Language Machine Learning
url	https://arxiv.org/abs/2504.15585