> ## Content Index > Fetch the complete content index at: https://www.theamericanquorum.com/llms.txt > Use this file to discover other available public pages before exploring further. # EU Court Adviser Says Google Gemini Could Infringe Press Publishers’ Rights - URL: https://www.theamericanquorum.com/eu-court-adviser-says-google-gemini-could-infringe-press-publishers-rights/ - Published: 2026-11-03T17:29:30.000Z - Updated: 2026-11-03T17:29:30.000Z - Description: An adviser to the EU’s top court said Google Gemini could infringe publishers’ rights when its answers recognizably reproduce news articles. The nonbinding opinion treats chatbot answers and model training separately. - Author: News Desk - Tags: Tech An adviser to the European Union’s highest court said Tuesday that Google’s Gemini chatbot could infringe press publishers’ rights when it retrieves articles from the web and uses them to answer users’ questions. The opinion draws a distinction between giving readers the facts from a news story and delivering an answer that recognizably reproduces the publisher’s work. First Advocate General Maciej Szpunar’s recommendation does not decide whether Google broke the law. It gives the Court of Justice of the European Union a framework for answering questions from a Hungarian court, which will still have to resolve the dispute. The case could shape how AI services handle material they retrieve at the moment a user asks for news. ## A dispute over how Gemini answers Like Company, a Hungarian operator of news websites, brought the case against Google Ireland. The publisher alleges that Google used articles from its sites in training Gemini and that the chatbot reproduced and made articles available to users. The Budapest court referred four questions to the EU court, dividing them between the training process and the answers that users receive, according to the court’s [case summary](https://curia.europa.eu/site/jcms/p1%5F1000088866/en/opinion-case-c-250/25-like-company?ref=theamericanquorum.com). That division matters because an answer about current events can draw on material that the model fetches after it was trained. In [developer documentation](https://aistudio.google.com/docs/google-search?ref=theamericanquorum.com), Google describes a Gemini feature that can search the web when prompted, process the results and supply links to sources. The documentation explains how that feature can work; it does not establish what happened when the Hungarian publisher’s articles were allegedly used. ## When an answer can substitute for the article Szpunar says the relevant publisher right protects the way an article expresses the news, while the underlying facts remain available for others to report. In its [press release](https://curia.europa.eu/site/upload/docs/application/pdf/2026-11/cp260155en.pdf?ref=theamericanquorum.com), the court said his analysis of reproduction looks at how the chatbot obtained the article, including whether it bypassed an effective barrier such as a paywall, and at what it returned to the reader. The key question about the answer is whether it makes the article’s expression recognizable to such a degree that a reader no longer needs the original. An answer could give the date or outcome of an event without recreating the article that reported it. An answer that preserves the article’s expression closely enough to take its place presents a different question under the proposed approach. The court summary also describes a separate concern about making material available to the public when related queries repeatedly produce answers with recognizable expression. The EU’s [copyright directive](https://eur-lex.europa.eu/eli/dir/2019/790/oj?ref=theamericanquorum.com) excludes hyperlinks, individual words and very short extracts from the press publishers’ right. That means the presence of a link and the contents of the answer are separate issues. A chatbot can point to a news story; the harder question is what else the chatbot displays beside that link and how it got the material in the first place. The adviser also addressed who could be responsible for the use. The court’s summary says his approach would hold providers liable where the conduct is built into their business model or they have failed to take credible and effective steps to prevent it. The judges have yet to say whether that is the standard they will apply. ## Training remains a separate question Szpunar recommended that the court decline to answer the two training questions in this case. He said the record did not show that the publisher’s articles were used to train Gemini. He nevertheless discussed how the EU text and data mining exception could generally permit the use of protected works for training, subject to its conditions. The directive’s general mining exception applies to lawfully accessible material and allows rights holders to reserve their rights in an appropriate way. Google told [Reuters](https://www.reuters.com/world/googles-use-press-by-ai-chatbot-may-infringe-copyright-rules-eu-court-adviser-2026-11-03/?ref=theamericanquorum.com) that the mining exceptions are important to AI development and welcomed the adviser’s recognition of a role for them in training. The [Law Society Gazette](https://production.lawsociety.ie/gazette/top-stories/2026/november/chatbots-may-be-infringing-press-rights/?ref=theamericanquorum.com) likewise described the opinion as addressing the live reuse of articles separately from the training allegations. The training discussion does not settle whether a particular chatbot response reproduced protected material. ## What the judges still have to decide The judges can adopt, narrow or reject the adviser’s approach. Their ruling will interpret EU law for the referring court, which will have to apply it to the evidence in the dispute. The opinion did not order changes to Gemini, grant damages or set a date for judgment. It also did not decide how United States copyright law would treat the same conduct. For publishers and AI developers, the practical issue is where the judges draw the line between a useful answer and one that substitutes for a news article. If they follow Szpunar, evidence of how a service accesses pages, how its answers reproduce material and whether safeguards work in practice could matter more than a general description of how the underlying model was trained. That is a question about the behavior of the product a reader uses, as well as the material that went into building it.