Does AI infringe copyright?

By Juha Vesala[*]University of Turku and IPR University Center.

1. Potential copyright infringing activities in AI development and use

Copyright infringement can occur at various stages of AI development and use. These include pre-training (e.g. collection and preparation of data for training), the training process itself, and the exploitation of the trained model (e.g. making available of the model). Indeed, it is necessary to consider specific activities associated with AI development and use to meaningfully answer the question whether AI infringes copyright by involving reproduction or making available of protected subject-matter. Moreover, the answer to these questions typically depends on the specific conditions in and means by which AI is developed or used (e.g. what data is used and how).

While infringement of subject-matter used to train an AI model is a central consideration at all stages of the AI lifecycle, from pre-training to use of AI models as part of services, also subject-matter that is otherwise sourced by AI-based services (e.g. subject-matter obtained from the web by an AI-based service) or provided by users (e.g. as prompts or uploaded files) can also be infringed when the use of AI is considered.