生成式人工智能数据“投喂”的著作权侵权行为规制  

Copyright Infringement Regulation of Training Data"Feeding"by Generative AI

作　　者：常烨[1] Chang Ye(Southwest University of Political Science&Law,Research Base of Legal System for Market Transaction,Chonqing 401120,China)

机构地区：[1]西南政法大学市场交易法律制度研究基地,重庆401120

出　　处：《科技与法律(中英文)》2025年第2期31-41,共11页Science Technology and Law(Chinese-English Version)

基　　金：中国法学会课题“营商环境法治化与商事法律制度现代化”(CLS(2022)D24)。

摘　　要：生成式人工智能在大模型训练中的数据“投喂”引发了内容创作者对自身作品被违法使用的担忧,生成式人工智能面对指控往往诉诸合理使用条款进行抗辩。生成式人工智能的技术特性,使得对于“投喂”未获许可作品行为的规制面临事实查明、归责、惩戒等多重困境。对于这种涉及新技术、新要素、新产业链的著作权侵权治理而言,必然对新制度有所诉求。因此,应当在生成式人工智能尚在自我探索的这一发展阶段,明确这种著作权侵权的法律适用,从侵权责任认定与举证责任分配、技术性授权的行业自律规范、“机器遗忘”强制性规范等多个维度,构建生成式人工智能输入端对数据“投喂”著作权侵权行为的规制框架。The data"feeding"of generative AI in large model training has caused great concerns among content cre-ators in illegal usage of their works,and generative AI often appeals to fair use terms to defend themselves against the charges.Due to the technical characteristics of generative AI,the regulation of"feeding"of unlicensed works faces multiple difficulties such as fact finding,imputation and punishment.As for this kind of copyright infringement regula-tion concerning new technology,new elements and new industrial chain,the demand for a new legal system will be in-evitable.Therefore,at the stage of self-exploration of generative AI,the essence is to clarify the law application con-cerning this kind of copyright infringement,to build up an ideal regulating frame of copyright infringement risk in data"feeding"at the input end of generative AI,and from multiple perspectives such as legal responsibility for tort,distribu-tion of proof burden,industrial self-discipline specification about technical authorization,mandatory norms about"ma-chine unlearning".

关 键 词：生成式人工智能 训练数据“投喂” 著作权 侵权 

分 类 号：D923.41[政治法律—民商法学]