生成式人工智能服务提供者的注意义务探析

Analysis of the Duty of Care for Providers of Generative Artificial Intelligence Services

摘要: 认定生成式人工智能服务提供者是否侵权，关键在于其是否尽到合理的注意义务。现行法律没有明确规定服务提供者的注意义务，学界也对此看法不一。本文采用过错责任原则，把注意义务分成了三阶段：风险预防、风险控制、风险处置。每个阶段各有三项义务。预防阶段是数据合法性审查、模型设计合规和用户管理。控制阶段是内容标识、风险提示和算法透明。处置阶段是投诉举报机制、内容阻断和动态修正。AI服务用户数量庞大、应用场景多样。AI的服务模式、应用场景、主体类型和用户群体等因素对注意义务都有影响，因素变化了，注意义务也会有所差异。该分类框架，有助于更客观地判断服务提供者是否有过错。

Abstract: Determining whether a generative AI service provider has infringed depends crucially on whether it has fulfilled its duty of reasonable care. Current laws do not clearly define the duty of care for service providers, and academic opinions remain divided on this issue. This article adopts the principle of fault liability and divides the duty of care into three stages: risk prevention, risk control, and risk response. Each stage entails three specific obligations. The prevention stage includes reviewing data legality, ensuring compliance in model design, and managing users. The control stage involves content labeling, risk warnings, and algorithm transparency. The response stage covers mechanisms for complaints and reports, content blocking, and dynamic adjustments. Given the vast number of AI service users and diverse application scenarios, factors such as service models, application contexts, types of entities involved, and user demographics all influence the duty of care—changes in these factors lead to variations in the required level of care. This classification framework helps achieve a more objective assessment of whether a service provider has acted negligently.