人工智能產品侵權歸責何以脫離產品責任？

人工智能產品侵權的歸責是適用產品責任，還是新設責任類型，比較法和國內法都無定論。基於受害者保護及促進產品安全的規範目的，當人工智能產品能滿足產品責任中產品及相關規則的隱含要求時，產品責任能容納該部分侵權的歸責。因人工智能技術內在的不可控，人工智能產品侵權必然會超出產品責任的容納範圍。對於超出部分的歸責，需在危險事由中確證其特殊性，即從抽象的一般性危險中具體界定符合人工智能產品致害特性的危險，從而有別於產品責任的危險和高度危險。這種人工智能產品致害危險源於生產者，基於非相互性風險理論，應由其承擔無過錯責任。在體系建構上，宜采以單行法新設責任類型與改進產品責任並行的方式。建構歸責事由合理、規則完善的人工智能產品侵權責任體系，既能對受害者提供充足保護，也能為人工智能技術的發展與運用留有寶貴空間。

The attribution of liability for artificial intelligence (AI) product-related infringements remains unsettled in both comparative and domestic law, with no definitive consensus on whether to apply product liability or establish a new category of liability. Considering the regulatory objectives of protecting victims and promoting product safety, product liability can accommodate AI-related infringements when AI products meet the implicit requirements of “products” and corresponding rules under product liability law. However, due to the inherent uncontrollability of AI technology, certain infringements will inevitably exceed the scope of product liability. For these cases, liability attribution should be grounded in the identification of AI-specific risks within the category of hazardous causes. This requires distinguishing AI-related risks from general product liability risks and highly hazardous risks by concretely defining the harm-causing characteristics of AI products. Since the source of such AI-generated risks originates from the producer, the principle of non-reciprocal risk theory supports imposing strict liability on them. From a structural perspective, a dual approach is advisable: simultaneously introducing a new category of liability through standalone legislation while refining the existing product liability framework. Establishing a well-structured AI product liability system with clearly defined attribution grounds and comprehensive regulations would not only ensure adequate protection for victims but also provide room for the advancement and application of AI technology.