Recently, the Court of Justice of the European Union (CJEU) has taken up its first-ever copyright case involving artificial intelligence. The case originates from Hungary and is titled Like Company v. Google Ireland Ltd, Case No. C-250/25.

The plaintiff, Like Company, is a news publishing agency that published an article on one of its protected online news sites, balatonkornyeke.hu. The article reported that renowned Hungarian singer Kozsó had not given up on his dream of keeping dolphins in an aquarium by Lake Balaton, Hungary’s largest lake. The article also detailed Kozsó’s hospitalization, personal interests, time served in the United States, and a fine he received for electricity theft.

However, the defendant’s AI chatbot, Google Gemini, generated a detailed response when prompted with the question: “Can you provide a Hungarian summary of the online news publication on balatonkornyeke.hu about Kozsó's plans to introduce dolphins into the lake?” The chatbot’s answer included summaries of information that appeared in the plaintiff’s publication.

The plaintiff claims that the chatbot’s output copied protected parts of its article and made them publicly available, thus infringing copyright. Like Company is seeking damages under EU law.

The CJEU has been asked to answer the following four questions:

Question 1 (with two sub-questions):
Should Article 15(1) of Directive (EU) 2019/790 (the Copyright in the Digital Single Market Directive) and Article 3(2) of Directive 2001/29/EC (the Information Society Directive) be interpreted as meaning that:

• • When a chatbot based on a large language model (LLM) generates a response that includes text identical to parts of a news publication (with a length qualifying for protection under Article 15 of Directive 2019/790), this constitutes a “communication to the public”?
  • If so, is the fact that the chatbot generates the reply using a technical process of predicting subsequent words based on language patterns legally relevant?

Question 2:
Should the above-mentioned provisions be interpreted as meaning that the process of training an LLM (through recognizing and learning language patterns) itself constitutes an act of “reproduction”?

Question 3 (assuming an affirmative answer to Question 2):
If training constitutes reproduction, does this activity meet the conditions for exemption under Article 4 (the Text and Data Mining, or TDM, exception) of Directive 2019/790?

Question 4:
When a user inputs a prompt into an LLM chatbot that matches or refers to the text of a news publication, and the chatbot subsequently generates a reply containing all or part of that publication’s content, does the service provider thereby commit an act of “reproduction”?

In response to the infringement allegation, Google asserts that Gemini does not store articles or retrieve static content, but instead dynamically generates outputs based on probabilistic predictions.