Brussels (Belgium) (AFP) – Artificial intelligence conversational agents can “in certain circumstances, infringe the rights of press publishers”, estimated Tuesday the Advocate General of the Court of Justice of the European Union, who will decide in around six months.

Published on: 03/11/2026 - 12:39Modified on: 03/11/2026 - 13:04

The Court is not bound by the conclusions of the Advocate General, but his opinions are most of the time followed.

This case - a dispute in Hungary between an online media operator and the American giant Google - is the first case relating to generative AI brought before the CJEU.

On Tuesday, Advocate General Maciej Szpunar delivered his conclusions, before a judgment expected within six months.

He called on the Court to “find that the retrieval and reuse, in real time, of online press publications by conversational agents (chatbots) based on artificial intelligence (AI)” can “in certain circumstances, infringe the rights of press publishers”.

In this case, the Hungarian online media operator, Like Company, accuses Google of having used its articles to train the language models (LMM) of its conversational assistant Gemini and of having reproduced and made them available to the public on several occasions via Gemini between June 13, 2023 and May 31, 2024.

Like Company claims copyright protection for its articles.

Hungarian justice decided not to rule at this stage and turned to European justice to ask legal questions.

In his conclusions delivered on Tuesday, the Advocate General of the CJEU considers that the training of a language model (LLM) on texts, images or other content protected by copyright is, in principle, legal in the EU without requesting authorization, thanks to the European exception known as "text and data mining".

Instead, he recommends focusing on the responses chatbots provide to users.

"Form of expression"

According to Maciej Szpunar, when a chatbot "retrieves a post from a news publisher's website, reuses that post to generate a response, and provides that response to the user, the news publisher's right may be violated in certain circumstances."

In his eyes, the problem does not concern the facts reported by the AI ​​conversational agent but "the form of expression" that it uses, if it directly takes up that of a media, to the point that the reader is "entirely exempt from accessing the original publication".

“It is a question of knowing, firstly, whether this conversational agent illicitly accessed the press publication in question, in particular by circumventing” paid access to a media.

And secondly to assess "whether the response provided to the user only reports facts contained in the original publication or, on the contrary, reuses the form of expression of this publication", considers the Advocate General.

After this opinion, a Google spokesperson emphasized that the Advocate General "recognizes that European Union law allows the training of AI models using publicly available information."

“Gemini learns” language models “in order to produce original content rather than reproducing existing works,” assures the American giant.

Google also highlights “numerous control mechanisms for publishers”, with “the possibility of opposing the use of their content”, says the company.

© 2026 AFP

The content you are trying to access does not exist or is no longer available.