This study examines the duty of care for AI software developers, focusing on copyright infringement issues raised by the proliferation of generative AI. Specifically, this study attempts to interpret the specific provisions of copyright law that may be infringed by AI software developers amid the proliferation of generative AI technology, distinguishing between the learning stage and the output and distribution stages.Currently, there is considerable confusion among Korean developers regarding the duty of care they must bear when developing generative AI software. This study examines the potential for copyright infringement at each stage of generative AI software technology and examines the potential legal issues that may arise. Next, this study explores the relationship between generative AI technology and copyright. To assist Korean developers in pursuing a sophisticated legal approach, this study explains the differences between traditional copyright law and the copyright requirements for generative AI technology. Therefore, to address the current anxiety surrounding the still significant uncertainty surrounding the ex post facto, fact-based, and expost facto duty of care for generative AI software developers, it is necessary to consider enacting a universal legal principle that can be summarized as follows: 1) only use legally accessible data for training; 2) minimize the removal or modification of Rights Management Information (RMI) and ensure transparency; and 3) control substantial similarity and market substitutability at the output stage. This represents a balance that protects the legitimate interests of copyright holders without hindering the development of AI technology. Going forward, our legal system should also gradually establish explicit regulations on TDM, transparency standards, and an accountability structure centered on output control, rather than relying solely on fair use interpretations to address the uncertainty surrounding AI learning. Such institutional reforms will lay the foundation for transforming generative AI technology from a “dangerous gray area” to a predictable and reliable area of innovation.
Park et al. (Sat,) studied this question.