生成式人工智能服务提供者侵权责任中的注意义务

Duty of Care in the Tort Liability of Generative AI Service Providers

摘要: 生成式人工智能的颠覆性发展对以人类行为为中心的侵权责任框架构成了根本性挑战。其侵权形态呈现出“人机协同”、“直接生成”与“算法黑箱”等全新特征，导致传统过错责任、无过错责任及“通知–删除”避风港规则均面临适用困境。注意义务作为联结过错认定与责任承担的关键枢纽，成为破解这一困局的核心。文章认为，生成式人工智能服务提供者并非传统网络服务提供者，而是兼具“守门人”与“信息生产者”角色的新型主体。其注意义务的理论来源可追溯至传播学“守门人”理论与侵权法危险控制理论，其正当性则建立在权利义务相一致、法经济学预防效率及全球数字治理趋势之上。在制度构建上，应摒弃非此即彼的归责模式，转向构建一个贯穿事前防范、事中干预、事后处置的全流程、分层级注意义务体系。具体而言，服务提供者需承担基于现有技术水平的同业高度注意义务，其判定基准应综合考虑预见可能性、控制力、获益情况及行业特性。通过合理设定注意义务，旨在技术创新与权益保护之间寻求动态平衡，引导人工智能产业向善发展。

Abstract: The disruptive advancement of generative artificial intelligence poses a fundamental challenge to the tort liability framework, which has traditionally been structured around human conduct. The distinctive modalities of infringement emerging in this context—namely, human-machine collaboration, direct generation, and algorithmic black boxes—render the existing doctrines of fault-based liability, strict liability, and the notice-and-takedown safe harbor regime substantially inadequate. The duty of care, serving as the pivotal nexus between the determination of fault and the imposition of liability, lies at the heart of resolving this predicament. This article contends that generative AI service providers are not conventional internet service providers; rather, they assume a hybrid role as both “gatekeepers” and “information producers.” The theoretical underpinnings of their duty of care can be traced to the gatekeeper theory in communication studies and the risk control theory in tort law, while its normative justification is grounded in the principle of congruence between rights and obligations, the preventive efficiency rationale of law and economics, and the imperatives of global digital governance. In terms of institutional design, a binary approach to liability allocation should be abandoned in favor of a comprehensive, multi-tiered duty-of-care system that operates across the entire continuum of pre-event prevention, in-event intervention, and post-event remediation. Concretely, service providers should be held to a high duty of care commensurate with the prevailing technological standards of the industry, the assessment of which should take into account foreseeability, degree of control, realized benefits, and sector-specific characteristics. By calibrating the duty of care appropriately, it is possible to strike a dynamic equilibrium between technological innovation and the protection of legal interests, thereby steering the development of the AI industry towards socially beneficial ends.