Abstract: The legality of generative artificial intelligence (GAI) development hinges on whether the act of copying vast amounts of texts during its training process constitutes copyright infringement. From the author-centered perspective， a work is seen as an extension of the author's personality or labor， entirely under the author's control. Unauthorized copying of a work by GAI infringes on the author's exclusive rights. Consequently， the traditional copyright infringement criteria have been based solely on “access + substantial similarity”. However， from the market-centered perspective， the author's copyright has market boundaries that do not extend to the areas of market failure beyond the primary market for the work. Therefore， the criteria for determining copyright infringement should be reconstructed， adding the element of “market harm” to the traditional “access + substantial similarity” framework. The focus should be on the impact of the work's use on its primary market. In the digital context， if GAI's massive copying of texts does not enter the market boundaries of the author's copyright， it cannot cause market harm and thus does not constitute infringement.

Key words: generative artificial intelligence (GAI); digital age; copyright infringement; copying behavior; dissemination behavior