Large Language Models and EU Intermediary Copyright Liability: Quo vadis?

Abstract

Large Language Models ( LLMs) are gaining a momentum. Their services to generate direct text or images on a topic in response to users’ prompts or produce textual summaries of a search inquiry skyrocketed their popularity. This, of course, does not come without challenges. A number of copyright infringement cases are now preoccupying the courts; amongst them, lawsuits about copyright infringing outputs produced by LLMs after users’ textual prompts. In light of the absence of any judicial guidance, this article aims to address how copyright infringement could be established in the context of outputs generated by LLMs and reflect on whether LLMs could be liable and if so what type of liability could be imposed; taking into consideration a line of case law from the CJEU and EU legal provisions. Finally, a set of novel recommendations that could accommodate the liability of LLMs for copyright infringement will be provided.