Chatbots and newspapers: EU Advocate General says real-time AI answers can infringe publishers' rights

On 3 November, Maciej Szpunar, Advocate General at the Court of Justice of the European Union, delivered his opinion in case C-250/25, which pits the publisher Like Company against Google Ireland. The headline of the Court's press release No 155/2026 sums up the proposal: when a chatbot retrieves and reuses online press publications in real time to answer users, it may, under certain circumstances, infringe publishers' rights. The system in question is Gemini, Google's chatbot formerly known as Bard. The preliminary reference was made by a Hungarian court in April 2025, with four questions according to Covington, and the text of the press release reproduced on the blogs of Stefano Quintarelli and IPKat describes it as the first generative AI case to reach the Court. The provisions at stake are two articles of Directive 2019/790: Article 15, on the related right of press publishers, and Article 4, on the text and data mining exception.

The opinion keeps two levels apart. On the two questions concerning training, Szpunar proposes declaring them inadmissible: according to the reported text, there is no evidence that Like Company's articles were used to train the Gemini models, and training an LLM does not amount to an “online use” of the publications, so the questions remain hypothetical. For the sake of completeness, according to the text reported by Quintarelli and IPKat, the Advocate General suggests that using protected works and other subject matter to train an LLM is, as a rule, permitted under the TDM exception. IPKat adds that the conditions of lawful access and the absence of a rights reservation still apply, and reports that for Szpunar Article 4 “could become a compensated limitation”. Again according to IPKat, the Advocate General rejected Google's argument that training is governed by the law of the place where the main computing infrastructure is located, and on this point asked the legislature for clarification.

On the answers, the passage reported by Quintarelli sets a precise test: the publisher's right to make its publications available to the public is, as a rule, infringed when their expressive form is “systematically recognisable” in the answers the chatbot gives to related questions. The provider is liable if such uses are inherent to its business model, or if it has not taken adequate measures to counter them credibly and effectively. According to IPKat, Szpunar considers it far from obvious that such uses fall under the exclusions for “mere facts”, “very short extracts” and “individual words”, rejects the idea that publishing online amounts to implied consent, and regards the TDM exception as inapplicable to answers. Since these are answers addressed to users in the EU, the same blog notes, the question of territoriality does not arise here.

A few limits should be kept in mind. The opinion, 240 paragraphs long according to IPKat, does not bind the Court, and no date has been given for the judgment. The official PDF of the press release could not be accessed: the sentences beyond the headline come from secondary reproductions, consistent with each other and with the headline published by the Court. The exact name of the Hungarian court that made the reference is not confirmed. The content of the article and of the Gemini answer that gave rise to the dispute has not been verified, and there is currently no official statement from Google on the opinion. What I find interesting is the shift in the centre of gravity: if the Court followed Szpunar, it would not rule on training, leaving only his indication, given for completeness, that training generally falls under the TDM exception subject to the conditions recalled by IPKat; the contest with publishers would be fought over the answers. And the second limb of liability, the one about adequate measures, moves part of the question from law to design: it may not be enough to say the model was not built to reproduce an article, because the provider would have to be able to show the measures it has taken, and would be liable anyway if those uses were inherent to its business model. For Italian publishers this is a case to follow closely, not least because the press release is available in Italian.

— Olya