摘要
生成式人工智能时代,既有隐私风险治理范式的局限性已经显现。大模型的技术新特质、广泛应用与二次应用使得其所引发的隐私风险,相较一般性隐私风险,呈现出开放性、级联性与隐蔽性特征。作为数据目的限制原则的拓展,模型目的限制原则通过将治理干预点从数据层面转向模型层面,强调模型用途的事前声明与技术全生命周期监管,为大模型隐私风险治理提供了一种新的进路。这一兴起于欧盟的大模型治理理论,在中国同样具备落地可行性,其可对当前主流的隐私风险应对进路形成有力补充,助力构建覆盖大模型全生命周期的数据层-模型层双重干预治理机制。
In the era of generative AI,the limitations of current privacy risk govermance approaches have emerged.The novel technological characteristics of large language models(LLMs),alongside their widespread and secondaryapplications,give rise to privacy risks that,compared with conventional privacy risks,are marked by greater openness,cascading effects,and concealment.As an extension of the data purpose limitation principle,the principle of model purpose limitation shifts the focus of governance intervention from the data layer to the model layer.It emphasizes the ex-ante declaration of model purposes and supervision throughout the entire technology lifecycle,providing a new possible approach to the governance of privacy risks of LLMs.Although this theory emerged in the context of EU LLMs governance,it also has practical applicability in the Chinese context.It can serve as a valuable supplement to current mainstream approaches to privacy risk mitigation and help build a dual intervention governance mechanism at both the data and model levels,covering the entire lifecycle of LLMs.
作者
王娟
WANG Juan(School of Literature and Media,Chaohu University,Hefei238024,Anhui,China)
出处
《昆明理工大学学报(社会科学版)》
2025年第6期49-58,共10页
Journal of Kunming University of Science and Technology(Social Sciences)
基金
安徽省社会科学创新发展研究课题青年项目“生成式人工智能时代人机间信息隐私鸿沟风险审视与识别应对研究”(2024CXQ525)
安徽省哲学社会科学规划课题青年项目“人工智能赋能安徽省级主流媒体系统性变革研究”(AHSKQ2024D108)
安徽省人文社科重点项目“智能传播时代安徽省新型主流媒体舆论引导力提升路径研究”(2023AH052076)。
关键词
大模型隐私风险
模型目的限制
数据目的限制
技术伦理
隐私对齐
privacy risks of LLMs
model purpose limitation
data purpose limitation
technology ethics
privacy alignment
