Experiences of Enforcing Copyright in a Generative AI Context

By Thomas Heldrup[*]Head of Content Protection & Enforcement, RettighedsAlliancen.

In my intervention I briefly described some of the experiences we have had in the Danish Rights Alliance enforcing copyrights in a generative AI context.

Looking first at the input stage of generative AI and training, I described how we discovered that a dataset comprised of approximately 200.000 illegal copies of e-books, sourced from an illegal German filesharing site, was collected and formatted with the aim of training generative AI models. The dataset named Books3 was made available for the public online with no restrictions.

I showed an example of how a book belonging to a Danish publisher was included in Books3.