GB/T 45652-2025Cybersecurity technology — Security specification for generative artificial intelligence pre-training and fine-tuning data (English PDF)

Foreword

This document is in accordance with the provisions of GB/T 1.1-2020 "Guidelines for standardization work Part 1: Structure and drafting rules for standardization documents" Drafting.

Please note that some of the contents of this document may involve patents. The issuing organization of this document does not assume the responsibility for identifying patents.

This document was proposed and coordinated by the National Cybersecurity Standardization Technical Committee (SAC/TC260).

This document was drafted by: Beijing Zhongguancun Laboratory, National Computer Network Emergency Response Technical Processing Coordination Center, China Electronics Technology Standards Institute of Chemistry, Peking University, Beijing Topsec Network Security Technology Co., Ltd., Beijing Kuaishou Technology Co., Ltd., Alibaba (Beijing) Software Services Co., Ltd., Beijing Baidu Netcom Technology Co., Ltd., Tsinghua University, Beijing Ruilai Smart Technology Co., Ltd., Tianyi Security Technology Co., Ltd.

Company, China Mobile Communications Group Co., Ltd., Xiaomi Technology Co., Ltd., Alibaba Cloud Computing Co., Ltd., Beijing Mianbi Intelligent Technology Co., Ltd.

Ltd., Hangzhou EZVIZ Software Co., Ltd., Beijing Institute of Technology, Beijing Zero One Everything Technology Co., Ltd., Institute of Automation, Chinese Academy of Sciences Institute, Lenovo (Beijing) Co., Ltd., Beijing Qihoo Technology Co., Ltd., iFlytek Co., Ltd., Huawei Cloud Computing Technology Co., Ltd.

Beijing Shuanxing Technology Co., Ltd., the Third Research Institute of the Ministry of Public Security, Ant Technology Group Co., Ltd., Beijing Venusstar Information Security Co., Ltd.

Technology Co., Ltd., Institute of Computing Technology, Chinese Academy of Sciences.

The main drafters of this document are: Xu Ke, Yao Long, Zhang Zhen, Liu Yong, Tan Zhixing, Li Qi, Xie Anming, Xu Xiaogeng, Yang Guang, Cui Tianyu, Hao Chunliang, Zhang Yanting, Xue Zhihui, Guo Jianling, Gu Chen, Jiang Wen, Ye Xiaojun, Tian Tian, Liang Wei, Jiang Weiqiang, Li Jiakun, Peng Juntao, Wang Huadong, Zheng Hongdong, Hong Yanqing, Wang Haitang, Zhu Guibo, Meng Yao, Zhang Xiangzheng, Liu Junhua, Li Fengfeng, Liu Yuhong, Liu Nan, Lin Guanchen, Wang Yan, Luo Hongwei, Tan Yingshui, Zhang Feng, Sun Xudong, Du Jinhao, Xu Shizhen, An Peng, Yu Yang, Sun Yong, Guo Jiexin, Wu Jianliang, Wang Xia, Wang Jinqiao, Gao Boya, Guan Ming, Wang Shijin, Zhao Lili, Wang Wenyu, Ding Zhiguo, Jiang Faqun, Sheng Qiang, and Wu Bowen.

Introduction

Pre-training and optimizing training data are the foundation of generative artificial intelligence and directly determine the quality and security level of generated content.

Pre-training and optimization of training data have security risks in the collection, pre-processing, and use of processing activities, and standards and specifications are urgently needed to improve pre-training Training and optimizing the security level of training data.

Cybersecurity Technology Generative AI Pre-training and optimize training data security specifications

1 Scope

This document specifies the security requirements for generative artificial intelligence pre-training and optimization training data and its processing activities, and describes the corresponding evaluation Price method.

This document applies to generative artificial intelligence service providers to carry out pre-training and optimization training data processing activities and security self-assessment.

It is also suitable for third-party organizations to conduct security assessments on pre-training and optimized training data.

2 Normative references

GB/T 35273

GB/T 41479-2022

3 Terms and definitions

The following terms and definitions apply to this document.