ICS35.030 CCSL80 T/CSAS 团 体 标 准 T/CSAS0029—2025 人工智能数据安全要求 Requirementsforsecurityofartificialintelligencedata 2025-11-28发布 2025-12-29实施 四川省网络空间安全协会  发布 全国团体标准信息平台 T/CSAS0029—2025 I目  次 前言·············································································································II 引言············································································································III 1范围···········································································································1 2规范性引用文件······························································································1 3术语和定义···································································································1 4概述···········································································································2 4.1人工智能生命周期组成················································································2 4.2人工智能数据活动主体及典型数据活动列举························································3 4.3常见安全风险···························································································3 5基本安全要求································································································3 6数据收集·····································································································4 7数据预处理···································································································4 8模型训练·····································································································5 9模型评估·····································································································5 10模型部署····································································································6 11模型应用····································································································6 12模型更新····································································································6 附录A（规范性）人工智能数据常见安全风险······························································8 附录B（规范性）语料及生成内容的主要安全风险·······················································10 B.1包含违反社会主义核心价值观的内容·······························································10 B.2包含歧视性内容·······················································································10 B.3商业违法违规··························································································10 B.4侵犯他人合法权益····················································································10 B.5无法满足特定服务类型的安全需求··································································11 参考文献········································································································12 全国团体标准信息平台 T/CSAS0029—2025 II前  言 本文件按照GB/T1.1—2020《标准化工作导则第1部分：标准化文件的结构和起草规则》的规定 起草。 请注意本文件的某些内容可能涉及专利。本文件的发布机构不承担识别专利的责任。 本文件由四川省网络空间安全协会提出并归口。 本文件起草单位：成都信息工程大学、成都理工大学、全域数据信息安全重点联合实验室西南实验 室、成都久信信息技术股份有限公司、泰和泰律师事务所、成都四方数安信息技术有限公司。 本文件主要起草人：白杨、刘珏麟、张露、李冬芬、多滨、李瑞、柯钦文、欧阳志强、罗荣、尹晓 东、陈福中、黄恒、卢志超（排名不分先后）。 全国团体标准信息平台 T/CSAS0029—2025 III引  言 为防范人工智能生命周期各阶段面临的隐私泄露、算法歧视、模型滥用等数据安全威胁，加强人工 智能技术全生命周期数据安全管理，特制定人工智能数据安全要求，旨在为人工智能研发机构、服务提 供方及监管部门提供全链条规范性指导。 全国团体标准信息平台 T/CSAS0029—2025 1人工智能数据安全要求 1范围 本文件规定了人工智能生命周期中数据收集、数据预处理、模型训练、模型评估、模型部署、模型 应用、模型更新各环节的数据处理活动的安全要求，重点约束存在数据安全风险的人工智能应用场景。 本文件所提的人工智能数据包括但不限于训练数据、评估数据、模型数据。 本文件适用于人工智能领域的数据处理者规范数据处理活动，监管部门、第三方评估机构对人工智 能数据处理活动进行监督、管理、评估参照使用。 2规范性引用文件 下列文件中的内容通过文中的规范性引用而构成本文件必不可少的条款。其中，注日期的引用文件， 仅该日期对应的版本适用于本文件；不注日期的引用文件，其最新版本（包括所有的修改单）适用于本 文件。 GB/T25069信息安全技术术语 GB/T35273信息安全技术个人信息安全规范 GB/T35274数据安全技术大数据服务安全能力要求术语 GB/T37988信息安全技术数据安全能力成熟度模型 GB/T39335信息安全技术个人信息安全影响评估指南 GB/T41479信息安全技术网络数据处理安全要求 GB/T43697数据安全技术数据分类分级规则 GB/T45574数据安全技术敏感个人信息处理安全要求 3术语和定义 GB/T25069—2022和GB/T35274—2023界定的以及下列术语和定义适用于本文件。 3.1 训练数据trainingdata 用于人工智能模型的输入数据样本子集。 [来源：GB/T41867—2022，3.2.34，有修改] 3.2 评估数据evaluationdata 用于评估最终人工智能模型性能的数据。 [来源：GB/T41867—2022，3.2.3，有修改] 3.3 投毒数据poisoningdata 在训练数据中被有意或恶意注入带有虚假、误导性或有害信息，使人工智能模型在后续推理或决策 中产生错误、偏见或恶意输出的数据。