The Doctrinal Landscape of Generative AI Evidence in China: Concepts, Classification, and Evidentiary Qualification

Abstract

Advances in generative artificial intelligence (AI) have extended machine involvement from identifying and predicting existing information to generating new text, images, audio, and video. Such materials are increasingly entering judicial fact-finding and raising new questions for established classifications of evidence and theories of evidentiary qualification. Through a thematic review of representative Chinese scholarship, this article examines the conceptualization of generative AI evidence, its relationship with adjacent categories of evidence, and its qualification as evidence. Existing studies approach the subject through the evolution of machine evidence, the operation of generative models, the internal classification of AI evidence, the evidentiary competence of generated information, and the composition of large language model materials. Although generativity has emerged as a relatively stable point of convergence, no consistent view has formed regarding the scope, evidentiary use, or superordinate category of such materials. Generative AI evidence overlaps with digital evidence, electronic data, big data evidence, algorithmic evidence, machine evidence, and scientific evidence in terms of medium, data scale, generation mechanism, the role of machines in its formation, and method of proof. Chinese scholarship addresses evidentiary qualification primarily through relevance, authenticity, reliability, lawfulness, and legally recognized forms of evidence, but remains divided over the generation process, human–machine collaboration, the allocation of technical risks between admissibility and the assessment of probative value, and the need for an independent evidentiary category. Future research should distinguish generated materials, generative tools, and technical materials used to examine AI systems, and should test existing theories against judicial cases and normative practice.