When an AI chatbot retrieves a press article in real time and reuses it to answer a user, the publisher’s rights “may, in certain circumstances” be infringed, Advocate General Maciej Szpunar advised the EU Court of Justice on 3 November, according to the court’s press release. The case, C-250/25, pits Hungarian publisher Like Company against Google Ireland over the Gemini chatbot, and the court calls it “the first generative AI case to reach the Court of Justice”.
The opinion is not binding. The court says its judges “are now beginning their deliberations in this case” and that judgment will be given at a later date. By our arithmetic, the opinion came 579 days after the Hungarian court’s request was lodged on 3 April 2025, and 238 days after the 10 March 2026 hearing.
The key insight: As we read it, the adviser draws the line at answer time, not at training: Advocate General Szpunar treats training as, in principle, covered by the text and data mining exception, while a chatbot that reuses press articles in real time may infringe publishers’ rights where it reaches them unlawfully or reuses their form of expression so recognisably that the user need not open the original, according to the court’s press release.
Update, 3 Nov: what the full opinion says. We have now read the full text of Advocate General Szpunar’s opinion, a provisional version published on EUR-Lex in English, its original language. It goes beyond the press release in four places; point numbers are the opinion’s own.
How the test may apply to Gemini. In “brief guidance” to the Hungarian court, Szpunar writes: “Subject to that court’s findings, it would appear that that is not the case”. He gives four reasons: that “Like Company did not put in place effective technological measures designed to restrict access to, or the scraping of, the articles published on its web portals”; that in its example “it is not apparent that the form of expression in all or part of the article concerned is recognisable in the answer”; that any such acts do “not appear to be inherent in Google’s businessmodel”; and that “Google seems to have put in place appropriate measures to mitigate the risk of such acts occurring” (point 239). This answers a question our piece listed as not established below.
Free access and consent.Google argued that a publisher leaving articles freely accessible “should be regarded as having implicitly consented” to their reuse by chatbots (point 224). Szpunar replies that “the putting in place of a paywall is a significant commercial choice”, which publishers may skip “because they prefer collecting advertising revenue from traffic more broadly, not because they implicitly consent” (point 228).
Training and the model. An output identical or near-identical to a work a model was trained on “may, in itself, constitute a reproduction of that work”, to be “determined on a case-by-case basis” (point 89). But he declines to treat the model itself as holding copies: that reading “would mean that LLMs are, in themselves, subject to the rightholders’ sole purview, irrespective of how they are designed and used” (point 92). On opt-outs, a reservation of rights counts where “state-of-the-art robots and associated technologies are capable of ‘understanding’ it with a sufficient degree of reliability at the moment of access” (point 148).
The opinion is not binding and the Court of Justice can depart from it; the facts are for the Budapest court to establish.
Like Company runs several web portals and claims protection for their articles under the press publishers’ right in Article 15(1) of the 2019 Copyright in the Digital Single Market Directive, the press release says. It argues that Google used the articles to train the Gemini models and “repeatedly reproduced and made them available to the public through the Gemini chatbot between 13 June 2023 and 31 May 2024”. That is a 354-day window, counting both end dates, by our arithmetic.
The Budapest Regional Court asked the Court of Justice four questions. Two concern training; two concern the answers the chatbot gives users. Szpunar advises the court to declare the two training questions inadmissible and to focus on the two about answers, “as these are relevant in the main proceedings”.
The press release describes Gemini as drawing on large language models that use “Retrieval-Augmented Generation”, or grounding, to collect information in real time from the web. To answer news queries, it says, the chatbot “often retrieves (or ‘scrapes’) press articles available online”.
Stages of Case C-250/25, Like Company v Google Ireland, in days: the request for a preliminary ruling was lodged on 3 April 2025, the hearing was held on 10 March 2026 and the Advocate General delivered his opinion on 3 November 2026; Like Company alleges use between 13 June 2023 and 31 May 2024. Day counts are our arithmetic. The opinion is not binding and the judgment will be given at a later date.
The Two-Factor Test for Chatbot Answers
Szpunar starts from a limit on the right itself. The release says it does not protect “the mere facts reported in press publications”, only their specific form of expression, because its aim is “not to give publishers a monopoly over the news itself”.
Whether a chatbot answer falls under the publisher’s reproduction right then depends on two factors, in his view. The first is whether the chatbot accessed the article unlawfully, “in particular by circumventing effective technical measures (such as a paywall)” that restrict access to, or web-scraping of, the publication.
The second is whether the answer merely reports facts or instead reuses the article’s form of expression “in a recognisable manner, to such an extent that the user, by reading that answer, is entirely dispensed from accessing the original publication”.
For the separate right of making available to the public, the release says that right “is, as a rule, infringed where the form of expression of the press publication at issue is systematically recognisable” in the chatbot’s answers to related queries.
Who Carries the Liability
The opinion also addresses who answers for it. The chatbot provider is liable, the release says, where such uses “are inherent to the business model of that provider” or where it “has failed to put in place appropriate measures in order to counter those acts credibly and effectively”.
That places the test on how an answer engine is built and run, not only on a single output. A provider that lets answers systematically reproduce recognisable press text, or that gets past a paywall to fetch it, is the case the adviser describes as infringing; a provider that summarises facts from lawfully reached pages is not, on this reading of the release.
Training: Inadmissible, With a Pointer
On the two training questions, Szpunar says they are hypothetical: there is “no evidence that Like Company’s articles have been used to train the Gemini LLMs”, and training an LLM “is not a form of ‘online use’ of press publications”.
He still gives a view “for the sake of completeness”: that using protected works and other subject matter to train an LLM is, “as a rule, freely permitted under the text and data mining exception”, the Article 4 exception in the same directive. So the opinion splits the four questions in half: training is pointed to the exception, while real-time answers are where infringement can arise.
Business Pill · RAG Explained
An evergreen Business Engineer explainer on retrieval-augmented generation: how a model answers from documents it fetches. It teaches the general idea only and does not discuss this case, Google or press publishers.
As we read the release, the two factors map onto two engineering choices: how a chatbot reaches a page (the release names circumventing technical measures such as a paywall) and how closely its answer follows the article’s wording.
As we read it, the liability test turns on the businessmodel: the release says the provider is liable where such uses are inherent to that model or where it has not put appropriate counter-measures in place.
Court of Justice of the EU, press release No 155/26, 3 November 2026
“the press publisher’s right does not protect the mere facts reported in press publications, but only the specific form of expression of those publications”
Three Implications
FOR AI ANSWER ENGINES The release says the provider is liable where the uses are inherent to its business model or where it has failed to put appropriate measures in place.
FOR PRESS PUBLISHERS The release says the right protects the specific form of expression of a publication, not the facts it reports.
FOR MODEL TRAINING The adviser proposes that the two training questions be declared inadmissible, and gives the view that training is, as a rule, freely permitted under the text and data mining exception, the release says.
The framework describes the governance layer this way: “The governance layer sits at the intersection of capability, regulation, national security, access control, and geopolitical alignment. It is where decisions are made about who gets access to frontier intelligence, under what conditions, and for which purposes.”
As we read it, the opinion applies the same question in reverse: under what conditions an AI system may access and reuse a publisher’s content. The framework does not discuss copyright, press publishers or this case.
What Is Not Established
We read the court’s press release and its case file; we did not read the full opinion text, which the release says is published on the CURIA website on the day of delivery. The release is described by the court as an unofficial document for media use. It does not say how the Hungarian court should apply the test to the Gemini answers at issue, and it gives no finding on whether Google circumvented any paywall.
The Court of Justice can depart from the opinion, and the release gives no date for the judgment. We did not contact Google or Like Company.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
In the first generative AI case at the EU Court of Justice, Advocate General Szpunar advises that a chatbot reusing press articles in real time may infringe publishers’ rights where it reaches them unlawfully or reproduces their expression so recognisably that users need not read the original, with the provider liable when that use is inherent to its businessmodel. He treats training as, in principle, covered by the text and data mining exception. The opinion is not binding, and 579 days after the referral, by our arithmetic, the judgment is still to come.
A note on sourcing. The opinion’s content is from the Court of Justice’s press release No 155/26 of 3 November 2026, which the court describes as an unofficial document for media use; we did not read the full opinion text. Procedural dates are from the CURIA case file. Day counts are our arithmetic. We did not contact Google or Like Company.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
