A chatbot that fetches a news article and rewrites it for a user can infringe the publisher’s EU rights in two situations, says Advocate General Maciej Szpunar. His opinion in Case C-250/25, Like Company v Google Ireland, came out on 3 November 2026, according to the Court of Justice press release. It is advice to the judges, not a ruling.
One is unlawful access, such as getting around a paywall. The other is an answer that reuses the article’s wording so closely that the reader no longer needs the original.
What the case is about
Like Company, a Hungarian operator of web portals, sued Google in the Budapest Regional Court over Gemini, Google’s chatbot. It argues that Google trained Gemini on its articles and reproduced them through the chatbot “between 13 June 2023 and 31 May 2024”. The Hungarian court sent four questions to Luxembourg: two on training, two on the answers.
The right at stake is narrow. Article 15(1) of the Copyright in the Digital Single Market Directive covers “the online use of their press publications by information society service providers”. It “shall not apply in respect of the use of individual words or very short extracts of a press publication”, and it does not cover hyperlinking.
Training is not this case
Szpunar advises the court to declare the two training questions inadmissible. He sees no evidence that Like Company’s articles trained Gemini, and he says training is not an “online use” of a press publication. For completeness, he adds that training is, as a rule, “freely permitted under the text and data mining exception”.
That exception is Article 4 of the same directive. Article 4(3) adds a condition: the use must not have been “expressly reserved by their rightholders in an appropriate manner, such as machine-readable means”. The opt-out still matters.
The two-part test for answers
The live issue is retrieval-augmented generation (RAG): the chatbot fetches pages in real time and writes from them. The first factor is whether it accessed the article “unlawfully, in particular by circumventing effective technical measures (such as a paywall)”. The second is whether the answer reuses its expression so that “the user, by reading that answer, is entirely dispensed from accessing the original publication.”
The provider is liable where such uses are “inherent to the business model” or where it failed to take measures “to counter those acts credibly and effectively”.
That bar is high. In a first reading for the Kluwer Copyright Blog, João Pedro Quintais notes Szpunar’s provisional view that the example before the Budapest court does not appear to meet it.
What happens next
The judges are deliberating and will rule “at a later date”. Reuters, in a report carried by Mint, says such opinions are usually followed, but the court can depart from them. Quintais lists two more Hungarian referrals on generative AI waiting: C-806/26 and C-829/26.
The split between facts and expression echoes the US. A judge said music publishers suing Anthropic had not shown one infringing Claude output. Access is the other lever, and it was at the center of Meta’s paused terms on scraping linked sites.
What to do this week
If your team runs an internal tool that summarizes news, check it against this test. It should not get around paywalls, it should honor machine-readable opt-outs, and its answers should state facts and link the source rather than rebuild the article.
If you publish, put your text and data mining reservation in machine-readable form and keep paywalls technical. Nothing changes legally until the court rules, so treat this as a direction, not a deadline.
Sources
- Court of Justice of the EU, press release No 155/26 on the opinion in Case C-250/25 (3 November 2026)
- Court of Justice of the EU, case summary and video for C-250/25 Like Company
- Directive (EU) 2019/790, Article 15 on press publications
- Directive (EU) 2019/790, Article 4 on text and data mining
- João Pedro Quintais, Kluwer Copyright Blog (3 November 2026)
- Reuters via Mint (3 November 2026)
Get the weekly email
The week's most useful AI news, a tutorial and an interview.
Martin Scott
Martin Scott is the editor for business and policy at AI Observer, covering startups and deals, regulation, health and science, and AI agency profiles. He follows the money and the legal text, and separates what a rule says from what people claim it says. His style is analytical and plain-spoken.
Spotted a mistake? Tell us on the corrections page.
