军事知识服务平台
合成数据的"合成"风险与法律规制
摘要 作为人工智能领域的"新燃料",合成数据依托数据脱敏和多维特征重组的技术路径,既解决了传统数据应用的隐私合规困境,亦突破了跨域协作的信息壁垒,实现了数字治理从数据资源化向数据工程化的范式跃迁.然而,既有的数据治理范式尚无法有效识别和规制合成数据全生命周期中呈现的溯源性风险、过程性风险和衍生性风险,数据仿真度的提升与特征维度的扩展客观上导致主体资格认定、客体权属界定以及应用场景规制等法律关系的复合性异变.治理的目标不再是单纯"管理"既有的数据,而是"塑造"合规的数据集.为此,需重构以结果为导向治理中的"数据控制者—处理者"二元架构,转向原始数据控制者在初始训练集构建阶段、算法开发者在模型优化过程中和运营方在应用场景拓展时的三元责任分配模态,实现合成数据从生成到传播再到价值实现的全链条治理.
Abstract:
Hailed as the"new fuel"for artificial intelligence,synthetic data relies on techni-cal approaches such as data anonymization and the recombination of multi-dimensional features.It not only addresses challenges in privacy compliance associated with traditional data use,but also reduces barriers to cross-domain data sharing and collaboration,thereby enabling a paradigm shift in digital governance from data resourceization to a data engineering paradigm.However,existing data governance frameworks have not yet effectively identified or regulated the traceability risks,process-based risks,and downstream risks presented throughout the lifecycle of synthetic data.As the fidelity of synthetic data increases and the dimensionality of features ex-pands,legal relationships become increasingly complex,particularly with respect to the determi-nation of legal status,the allocation of rights in data,and the regulation of use contexts.The ob-jective of governance is no longer merely to manage existing datasets,but to shape datasets that are legally compliant by design.To this end,the conventional binary model of"data con-troller-data processor"in result-oriented governance should be reconfigured toward a tripartite allocation of responsibilities:the original data controller at the stage of initial training dataset construction,the algorithm developer during model development and optimization,and the opera-tor in the deployment and expansion of use contexts.This shift enables end-to-end governance of synthetic data across its lifecycle-from generation to dissemination and ultimately to value real-ization.
刊 名 东方法学 ISTICPKU
年,卷(期): 2026, (2)
分类号: DF31
机标分类号: D920.4 TP391 F276.6
在线出版日期: 2026年6月3日
基金项目: JS2025ZSPY0060:中央高校基本科研业务费专项资金资助项目,