Legal Issues Regarding Licensing of Training Datasets and Trained Models

Abstract

 The Contract Guidelines on Utilization of AI and Data by METI assume that appropriate negotiations are conducted in advance between the parties concerned regarding the licensing of data, etc. that do not fall under the category of copyrighted works. However, in practice, OSS licenses are often used to distribute training datasets and trained models that are not copyrighted works. It is understood that a license using an OSS license is valid when it is not immediately clear whether the object of the license is a copyrighted work. It is reasonable to interpret “use” in the license terms as equivalent to the “use” under the Copyright Act, even for data that is not a copyrighted work. The issues of commercial and for-profit use and so-called ShareAlike can also be considered on the premise of such an interpretation. It may be necessary to discuss the clause concerning the distribution of data that does not fall under the category of copyrighted works in an international data protection forum.