论生成式人工智能服务提供者的注意义务  

On the Duty of Care of Generative Artificial Intelligence Service Providers

作　　者：毕文轩 BI Wen-xuan(School of Law,Southeast University,Nanjing Jiangsu 211189,China)

机构地区：[1]东南大学法学院,江苏南京211189

出　　处：《苏州大学学报(哲学社会科学版)》2025年第2期90-103,共14页Journal of Soochow University(Philosophy & Social Science Edition)

基　　金：2024年中国法学会民法学研究会青年学者研究项目“生成式人工智能服务提供者注意义务研究”(项目编号:2024MFXH010)的阶段性成果。

摘　　要：生成式人工智能服务提供者由于其特殊的运行机制并不属于传统的网络服务提供者或者网络内容服务提供者,应当属于一种新型网络服务提供者。传统的“通知删除”规则以及“红旗规则”对其所导致的侵权行为规制力度有限,且存在较大的适用困境。对于生成式人工智能服务提供者注意义务而言,其来源包括立法规范、合同义务以及行业规范,而危险控制理论以及利益平衡理论可以作为正当性依据。在此基础上,生成式人工智能平台注意义务的判定基础应当从主客观两方面展开。在客观层面,需要考虑生成式人工智能服务提供者的对侵权后果的可预见性、近邻性以及公平正义的政策考量;在主观层面,需要围绕生成式人工智能技术的具体实施情况重构“理性人”标准。在实践中,应从风险预防、风险控制以及风险处置三个维度,构建生成式人工智能服务提供者注意义务履行的具体措施:在风险预防层面,生成式人工智能服务提供者需要履行数据合法性审查义务,模型设计的合法性审查义务,生成内容的预防控制义务,以及用户行为的规范管理义务;在风险控制层面,生成式人工智能服务提供者需要履行披露算法决策的基本逻辑义务,披露生成内容的来源和使用规则义务,以及披露生成模型的局限性和潜在风险义务;在风险处置层面,生成式人工智能服务提供者需要履行建立健全投诉机制和及时反馈义务,对生成内容的及时处理义务并实施合规性审查与动态修正模型的义务。Generative AI service providers,due to their unique operational mechanisms,do not fall into the traditional categories of network service providers or content service providers.Instead,they represent a new type of network service provider.Conventional regulatory frameworks,such as the“notice-and-takedown”rule and the“red-flag”principle,are limited in their effectiveness and present significant challenges in addressing infringements caused by generative AI.The duty of care for generative AI service providers is derived from legal norms,contractual obligations,and industry standards,with theoretical justifications based on principles of risk control and balancing of interests.Objectively,it requires an assessment of foreseeability,proximity,and policy considerations of fairness and justice regarding the potential consequences of the breach.Subjectively,the“reasonable person”standard must be reconstructed based on the specific implementation of generative AI technologies.In practice,the fulfillment of the duty of care should encompass three dimensions:risk prevention,risk control,and risk mitigation.For risk prevention,providers must ensure the legality of data,model design,and content generation,as well as regulate user behavior.In risk control,they must disclose the logic behind algorithmic decisions,the sources and usage rules of generated content,and the limitations and potential risks of their models.To mitigate risk,providers are required to establish effective complaint mechanisms,promptly deal with infringing content,conduct compliance reviews,and dynamically adjust their models.

关 键 词：生成式人工智能 服务提供者 注意义务 分层建构 

分 类 号：D922.16[政治法律—宪法学与行政法学] D922.17[政治法律—法学]