| 刘佳琨.大模型训练中已公开个人信息处理规则的适用困境及其纾解[J].南京邮电大学学报(社会科学版）,2026,(04):40~52 |
| 大模型训练中已公开个人信息处理规则的适用困境及其纾解 |
| The dilemma and resolution of rules on processing publicly disclosed personal information in LLM training |
| |
| DOI： |
| 中文关键词: 大模型训练 人工智能 已公开个人信息 个人信息处理 个人信息保护 |
| 英文关键词:LLM training artificial intelligence publicly disclosed personal information personal information processing reasonable scope |
| 基金项目:国家社会科学基金重大招标项目（ 20&ZD179） |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 已公开个人信息因获取便利而被广泛应用于大模型训练，但其合法性边界尚不清晰。大模型训练所具有的目的泛化、过程不可逆与风险形态衍生等特征，导致现行《个人信息保护法》中以 “合理范围 ”“重大影响 ”和拒绝权等为核心的已公开个人信息处理规则在适用中面临不确定性，“告知—同意 ”和删除机制难以有效发挥作用。结合大模型训练的技术特征与实践需求，可以从来源场景、处理方式、聚合风险等层面细化已公开个人信息合理处理的判定标准，通过技术措施、输出控制和行业规范强化对个人信息权益的过程性保障，促进个人信息保护与人工智能发展相协调。 |
| 英文摘要: |
| he training of Large Language Models（LLMs） depends extensively on large. scale datasets，and publicly disclosed personal information is frequently used due to its accessibil. ity，while the legal boundaries governing such use remain insufficiently defined. The characteris. tics of LLM training—namely purpose generalization，process irreversibility，and the derivative nature of associated risks—give rise to uncertainty in the application of the Personal Information Protection Law’s framework for processing publicly disclosed personal information，which is struc. tured around standards such as reasonable scope，significant impact，and the right to object. Con. sequently，traditional mechanisms based on notice and consent，as well as deletion rights，encoun. ter structural limitations in this context. Comparative approaches in other jurisdictions address these challenges by delineating clearer processing boundaries and by establishing exclusionary or exemption.based mechanisms to accommodate technological innovation. Taking into account the technical attributes and practical requirements of LLM training，the criteria for assessing the law.ful processing of publicly disclosed personal information can be further specified along dimensionsincluding source contexts，processing modalities，and aggregation.related risks. At the same time， procedural safeguards for personal information rights may be reinforced through technical mea. sures，output control，and industry standards，thereby facilitating a more coordinated developmentof personal information protection and artificial intelligence. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |