Abstract：As a key technology empowering the transformation of government intelligence in the era of digital intelligence,the application scenarios and capability boundaries of embedding large models into government systems continue to expand,promoting the transition from “Information as a Service” and “Platform as a Service” to “Model as a Service” government forms.Large models represented by DeepSeek have good localization deployment capabilities and the potential for “secondary development”. While customizing training and integrating applications for specific government scenarios,they also pose structural and multidimensional data security risks at the model deployment,training,and application levels.Specifically,it manifests as technical endogeneity and scenario associated data security risks at the model deployment level,training data sources and content security risks at the model training level,and explicit and implicit data security risks at the model application level.Fundamentally,the above-mentioned risks are deeply embedded in the government system's acceptance of the application of large model technology.Based on the TOE framework,a three-dimensional governance model of “technology reinforcement organizational collaboration environmental optimization” is constructed,proposing to build a diversified collaborative organizational system and an inclusive and prudent regulatory environment on the basis of strengthening technological autonomy and controllability,achieving a value balance between encouraging innovation and risk regulation.
王正超. 数智时代政务大模型数据安全风险及治理机制研究[J]. 中国科技论坛, 2025(11): 151-160.
Wang Zhengchao. Data Security Risks and Governance Mechanisms of Government Large Models in the Digital Intelligence Era. , 2025(11): 151-160.
2025, 
