論通知規則在生成式人工智能侵權中的適用

On the Application of Notification Rule in the Scenario of Generative AI Infringement

通知規則是歸責條款；而非免責條款。這決定了通知規則可以直接適用于生成式人工智能侵權；而非類推適用。反通知在生成式人工智能侵權中的缺失不足以成為否定通知規則可適用性的充分理由。通知中“準確定位侵權內容的信息”可通過對話記錄和截圖的方式提供；且通知人可以不是權利人。傳統通知規則中的“必要措施”在生成式人工智能侵權中發生了諸多變化；包括直接避免損害擴大的措施從處理已發生的侵權轉變為預防侵權再次發生；警示性措施從警示實施了侵權行為的網絡用戶轉變為警示生成式人工智能使用者；擔保性措施無法適用、限制乃至停止提供服務的措施從針對網絡用戶轉變為針對誘發致害內容生成的使用者。生成式人工智能服務提供者在收到通知後負有避免致害內容再次生成的義務；但該義務的邊界鬚根據個案中的具體商業模式、人工智能技術發展狀況等因素綜合判斷；且應有期限限制。在致害內容無法再次生成時；服務提供者負有告知義務。

The notification rule is a liability fixation clause, not a liability exemption clause in China. This determines that the notification rule can be applied to Generative AI infringements directly, rather than by analogy. The absence of counter-notification in the scenario of Generative AI is not a sufficient reason to deny the applicability of notification rule. The“information to accurately locate the infringing content”in the notification can be provided by the conversation records and screenshots, and the complaining party could not be the right holder. In Generative AI infringement cases, there are several changes in the“necessary measures”stated in traditional notification rule: (a) measures to prevent the expansion of damage shift from focusing on the infringing content that has already occurred to preventing infringing content in the future; (b) warning measures shift from alerting tortfeasors to alerting users of Generative AI; (c) security measures are not applicable; and (d) measures restricting or even ceasing shift from against network users to users who induce the generation of infringing content. Generative AI Service Providers are obligated to avoid the generation of infringing content after receiving the notification, but the boundary of the obligation should be determined according to the specific business model, the development of artificial intelligence technology, and other factors. Also there should be a time limited duration of this obligation. Generative AI Service Providers are obligated to inform the complaining part when the infringing content is failed to be generated again. Key words: generative artificial intelligence; notification