生成式人工智能场域下个人信息规范保护的模式与路径

Models and Paths for the Normative Protection of Personal Information in the Filed of Generative Artificial Intelligence

生成式人工智能场域下现有的个人信息规范保护机制失灵,引发个人信息规范保护的新问题,体现为对个人信息知情同意的判断困难以及对个人信息非法转移、过度处理与非法篡改等治理难题。而造成这些问题的原因是构建于生成式人工智能兴起前的个人信息保护规范采用赋权保护模式,难以满足生成式人工智能带来的动态化、差别化与风险控制的个人信息保护需求,因此应根据生成式人工智能阶段化的特点更新采用个人信息分阶保护模式,在生成式人工智能不同阶段采取类型化的个人信息保护路径:在初始预训阶段根据内部性特征确立基于权利的合规治理路径,在应用部署阶段根据场景化特征采取基于场景的风险控制路径,在结果生成阶段根据后果固定性特征采取基于后果的责任追究路径,在生成式人工智能全生命周期中实现对个人信息的有效规范保护。

The failure of the existing mechanism for the normative protection of personal information in the field of generative artificial intelligence has led to new problems in the normative protection of personal information,which is reflected in the difficulty of determining the informed consent for personal information and the governance problems of illegal transfer,excessive processing and illegal tampering and utilization of personal information.The reason for these difficulties lies in the fact that the personal information protection norms adopt the empowerment and protection model formed before the rise of generative artificial intelligence,which is difficult to meet the dynamic,differentiated and risk control needs of personal information protection brought by generative artificial intelligence.Therefore,according to the phased characteristics of generative Al,a phased protection model should be adopted,that is,different types of normative protection of personal information should be adopted according to the different stages of generative AI:a rights-based compliance governance path should be established on the basis of internal characteristics in the initial pre-training stage,a scenario-based risk control path should be adopted according to dynamic scenario-based features in the application deployment stage,and a consequence-based accountability path should be adopted according to the fixed characteristics of consequences in the result generation stage,to realize the effective and normative protection of personal information throughout the entire life cycle of generative artificial intelligence.