摘要
生成式大模型具有广泛的应用前景。大模型的训练和运行均需要海量数据的支撑,极有可能引发数据安全风险。认知风险是化解风险的前提,需要从静态、动态两个视角建立起大模型应用数据安全风险的认知体系。结合欧盟、美国等大模型的治理经验,针对我国大模型数据安全风险治理存在的不足,建议建立基于数据安全风险的分类监管路径、完善基于大模型运行全过程的数据安全责任制度、探索基于包容审慎监管的创新监管机制,为实现大模型应用的可信未来提供充分的法治保障。
Generative large models have a wide range of application prospects,however,the training and operation of those models need the support of massive data,which is very likely to cause data security risks.Cognitive risk is the premise of risk resolution,and it is necessary to establish a cognitive system of data security risk of generative model application from both static and dynamic perspectives.Combining the governance experience of generative models in the EU and the United States,and addressing the deficiencies in the governance of data security risks of generative models in China,it is recommended to establish a categorized regulatory path based on data security risks,improve the data security responsibility system based on the whole process of large model operation,and explore the innovative regulatory mechanism based on the inclusive and prudent regulation,in order to provide sufficient rule of law guarantee for realizing the credible future of large model applications.
作者
刘羿鸣
林梓瀚
Liu Yiming;Lin Zihan(Institute of Cyber Governance,Wuhan University,Wuhan 430072,China;Shanghai Data Exchange,Shanghai 201203,China)
出处
《网络安全与数据治理》
2023年第12期27-33,共7页
CYBER SECURITY AND DATA GOVERNANCE
