Liability Analysis of Generative Artificial Intelligence in Judicial Application: An Example of Generative Large Models

Abstract

The application of generative artificial intelligence technology in the judicial field has given rise to core legal issues, including disputes over subject qualification, the delineation of service attributes, and the distribution of tortuous liability. Taking typical cases such as the Tencent Dreamwriter case as the starting point, this paper systematically explores the difficulties in identifying the tortuous liability of generative artificial intelligence under the existing legal framework, focusing on disputes over the liability of service providers, the predicament of determining negligence caused by algorithmic flaws, and the ambiguity in defining infringing acts. Through a comparative analysis of domestic and international legislative practices, the study reveals the "instrumental" essence of generative AI and the core of disputes regarding its legal nature: as the controllers of the technology, service providers are obligated to fulfill legal responsibilities such as data compliance review, labeling of generated content, and filtering of illegal information. Furthermore, the study proposes a functionalist approach to determining tortuous liability, combining the "equivalent causation theory" with the principle of judicial classification to construct a dynamic framework for liability division. At the level of risk regulation, China has established a governance system integrating policy guidance and technical supervision, and has strengthened algorithmic transparency and content traceability through the Measures for the Labeling of Artificial Intelligence-Generated Synthesized Content. The purpose of this paper is to provide both theoretical support and practical guidance for the judicial discretion and institutional improvement in disputes related to generative AI infringement.