中圖分類號：TP13；TP309文獻標志碼：ADOI:10.19358/j.issn.2097-1788.2026.06.008中文引用格式：彭文華.數據集安全：內涵、挑戰與防護策略［J］.網絡安全與數據治理，2026，45（6）：57-65.
英文引用格式：Peng Wenhua.Dataset security:connotation,challenges,and protection strategies［J］.Cyber Security and Data Governance，2026，45（6）：57-65.

Dataset security:connotation,challenges,and protection strategies

Abstract： Drawing on multidomain case studies from government public services,medical surveillance,and military applications,it systematically analyzes security risks encountered during dataset collection,preprocessing,labeling,training/application,and modeloutput stages.By contrasting datasets security with traditional data security,this paper distills targeted prevention logics and concrete countermeasures.The research reveals that dataset security is an AIspecific extension of data security,currently facing unique risks such as data bias,poisoning attacks,and adversarial interference,with risks profiles varying significantly across sectors.Establishing a fullchain "technologymanagementcompliance" closedloop riskcontrol system for datasets can effectively enhance AI dataset security and provide both theoretical and practical guidance for the secure development of AI applications.

Key words : artificial intelligence; AI security; dataset security; dataset risk