The Court of Justice to rule for the first time on the interaction between generative AI and copyright: five things you need to know

The applicant is a Hungarian press publisher whose content appeared in reponses generated by Google's AI chatbot, Gemini. The claimant alleges unauthorised reproduction and communication to the public of press articles, via a chatbot-generated summary of a protected article about a local celebrity.

One central question: does the display, in a chatbot’s response, of a text fragment that closely matches protected press content amounts to communication to the public under Article 3 (2) of Directive 2001/29 and/or an infringement of the new press publishers’ right under Article 15 (1) of Directive 2019/790?

The Court is also asked to decide whether using protected content during the training of large language models (LLM) constitutes reproduction under Article 2 of Directive 2001/29. If so, the Court must clarify whether the text and data mining (TDM) exception under Article 4 of Directive 2019/790 applies and under what conditions. The question also arises whether training could be covered by the temporary reproduction exception in Article 5(1) of Directive 2001/29.

A key point of discussion is whether the chatbot’s outputs go beyond the permitted use of “individual words or very short extracts” of press publications under Article 15 (1), especially when the user’s prompt refers itself to a specific press content. The publisher argues that even short summaries should require explicit consent and compensation under Article 15 (1). Google, for its part, denies that the chatbot's outputs amount to reproductions or communications, especially considering how LLMs work.

The Court’s decision could have a decisive impact on how EU copyright law applies to AI. It could also redefine the limits of publishers’ rights and clarify how far AI developers can go when training models or generating content.