論生成式人工智能服務提供者過錯的認定

On Fault Identification of Generative Artificial Intelligence Service Providers

在人工智能生成內容致人損害的情形下，過錯是服務提供者承擔侵權責任的歸責事由。合理地認定服務提供者的過錯，是有效平衡預防風險與鼓勵創新雙重價值目標的關鍵。服務提供者過錯的本質是其違反了交往安全義務或注意義務。在認定標準上，應採取動態的“合理人”標準，結合技術水平、服務類型與侵權內容等維度，分析特定情境下合理的、謹慎的服務提供者應盡的注意義務。服務提供者的注意義務主要包括語料處理義務、對齊微調義務、內容審查義務、內容標識義務和用戶管理義務，若服務提供者未能合理地履行這些義務，則應認定為具有過錯，需對生成內容致人損害的行為承擔侵權責任。服務提供者侵權責任的認定應類推適用“通知規則”，這不僅有利於實現服務提供者與權利人之間的利益平衡，還可以減輕權利人證明服務提供者過錯的舉證負擔。

In the case of damage caused by the AI-generated content, the fault-based rule is the liability rule for generative artificial intelligence service providers to bear the tort liability. Reasonably identifying the fault of service providers is the key to effectively balance the dual goals of preventing risks and encouraging innovation. The fault of the service provider is essentially a breach of its duty of security or duty of care. In terms of the criteria, we should adopt the dynamic“reasonable man”standard to analyze the duty of care of a reasonable and prudent service provider under a specific situation considering technical level, service types, infringement content and other dimensions. Service providers' duty of care mainly includes duty of corpus processing, duty of alignment, duty of content review, duty of content labeling and duty of user management. If service providers fail to fulfill these duties reasonably, they should be deemed to be at fault and bear tort liability for the damage caused by the AI-generated content. The notice-and-takedown rule should be applied to the determination of tort liability of service providers, which is not only conductive to the balance of interests between service providers and right holders, but also can reduce the burden of proof for right holders to prove the fault of service providers.