Advocate General Maciej Szpunar proposes to the Court of Justice of the European Union (CJEU) in his Opinion of 3 November 2026: An AI chatbot can infringe the rights of press publishers if it retrieves their articles live and reuses them for its answers.
This is not binding: The Court of Justice will deliberate and deliver its judgment later. Case C-250/25 (Like Company) is the first case on generative AI to reach the CJEU.
What are Like Company and Google disputing?
Like Company, operator of Hungarian web portals, has sued Google Ireland before a court in the area surrounding Budapest. The claimant asserts the related right that EU law grants press publishers for the online use of their press publications (Article 15 of Directive (EU) 2019/790). According to its allegation, Google used the articles to train the Gemini models. Between 13 June 2023 and 31 May 2024, Gemini allegedly reproduced the articles multiple times and made them available to the public.
Gemini works with Retrieval-Augmented Generation (RAG, also called grounding): The chatbot fetches information live from the internet and often retrieves press articles in the process. The Hungarian court has referred four questions to the CJEU: two on the training of language models, two on the answers the chatbot gives to users.
What does the Advocate General recommend?
In Szpunar's view, the CJEU should reject the two questions on training as inadmissible. Evidence that the claimant's articles were used to train the Gemini models is lacking, and training is not an online use of press publications. For the sake of completeness, he notes: The use of protected works for training a language model is, in principle, freely permitted under the exception for text and data mining.
Whether an answer affects the publisher's reproduction right depends, in his account, on two factors:
- Access: Did the chatbot access the article unlawfully, for example by circumventing a paywall or other technical barriers against automated extraction (web scraping)?
- Reproduction: Does the answer only state facts from the article, or does the answer recognizably adopt its form of expression, to the extent that users no longer need to visit the original?
A chatbot generally infringes the right of communication to the public as soon as the form of expression of a press article is systematically recognizable in its answers. Under Szpunar's proposal, the provider is liable if such uses are an integral part of its business model or if it fails to take appropriate measures to counter them credibly and effectively.
What does the press publishers' right protect and what does it not?
The right protects the specific form of expression of a publication and the investments that went into creating it. Reported facts are not covered, and the right does not give publishers a monopoly on the news itself.
What the press publishers' right protects
How Advocate General Szpunar interprets the right in his Opinion of 3 November 2026
- The specific form of expression of a publication
- The investments in creating the publication
- The reported facts
- A monopoly on the news itself
The Opinion is not binding on the Court of Justice
Source: Court of Justice of the European Union
Szpunar's proposal separates two processes that often blur together in the debate: the training of a model and the individual answer. In his reading, the risk for providers lies with chatbots that retrieve current articles and recognizably repeat their wording in the answer.
What this means for you
The proposal changes nothing about current law. The Court of Justice's press release deals with the liability of chatbot providers; it says nothing about users. Nevertheless, the Advocate General's yardstick works as a rule of thumb for your work:
- It's best to read the original before you use a chatbot answer on current news. Ask the chatbot for its sources and write down the facts in your own words, with a link to the article.
- It's best not to adopt passages in which wording from a press article recognizably recurs, for example in newsletters or client texts.
- Make a note of the case number C-250/25 and check after the judgment whether your practice needs to change for texts you research for.
The lesson Copyrights and Copyright Law in the Gemini course provides the basics. Another recent report also focuses on AI training: A robots.txt entry is meant to exclude AI training at Google and Microsoft.
The Court of Justice published the Opinion on 3 November 2026.
Sources
