An EU court adviser has warned that Gemini copyright exposure can arise when Google’s chatbot retrieves press articles and reuses their wording in answers. Advocate General Maciej Szpunar told the Court of Justice of the European Union on 3 November 2026 that the press publishers’ right may be infringed in certain circumstances, while training is generally covered by a statutory exception. It comes in the first case in which the EU’s top court is examining generative AI and copyright, and it is not itself a ruling.
Key Takeaways
- Szpunar said Gemini answers may infringe press publishers’ rights in some cases.
- Training on protected text is generally allowed under the mining exception.
- Risk rises if a chatbot bypasses paywalls or copies recognisable wording.
- The opinion is not binding; judges usually follow advocate general advice.
What the Gemini Copyright Case Asks
Hungarian publisher Like Company sued Google Ireland, which operates Gemini as part of the Alphabet group, after the chatbot summarised its articles. The company runs several news portals and alleges Google used its stories to train large language models, then reproduced them in answers between June 2023 and May 2024. A Budapest court referred four questions in April 2025: two on training, two on answers drawn from articles online. The reference is case C-250/25.
Like Company argues the chatbot reproduced its content without permission, going beyond short extracts, The Next Web reported. Google has argued that Gemini does not memorise articles, that potentially infringing outputs would fall under exceptions for temporary copies and text and data mining, and that Hungarian law should not apply because Gemini was not trained in Hungary.
Training and the Text-and-Data-Mining Exception
Szpunar proposes that the training questions be set aside. He finds no evidence in the file that Like Company’s articles were used to train Gemini, and he does not treat that training as “online use” under the neighbouring right for press publishers, Paperjam reported.
If the court examines the point anyway, he considers training on protected works is, “as a general rule”, permitted under the text-and-data-mining exception. The Straits Times, citing a court statement, quoted him calling that use “freely permitted under the text and data mining exception”. He added that copyright rules do not protect “mere facts”, but “only the specific form of expression”.
When Live Retrieval Crosses the Line
The assessment changes when Gemini searches for an article in real time. Paperjam described the method as retrieval-augmented generation: the chatbot gathers current pages from the web rather than relying only on training data. Szpunar’s view, in the court statement carried by The Straits Times, was direct:
When an LLM (large language model)-driven chatbot retrieves a publication from a press publisher’s website, reuses that publication to generate an answer, and provides that answer to the user, the press publisher’s right may be infringed in certain circumstances.
He treated that risk as two conditions that can combine:
- The chatbot circumvents an effective technical measure, such as a paywall or a control designed to prevent automated scraping.
- The answer reuses the publication’s form of expression in a recognisable manner, so the reader no longer needs the original.
Morningstar, citing the court’s adviser, reported the second test as reuse “to such an extent that the user, by reading that answer, is entirely dispensed from accessing the original publication.” Neighbouring rights protect expression and the investment in producing the publication, not a monopoly over what was reported. Systematic reproduction of that wording may also amount to making the work available to the public.
The user who posed the query would not be liable. The provider could be, where reuse of posts is inherent to its business model or where it has not put “credible and effective” measures in place to prevent it.
| Operation | Proposed treatment |
|---|---|
| Training on protected text | Generally permitted under the text-and-data-mining exception |
| Restating reported facts | Not covered; the right protects expression, not mere facts |
| Retrieval plus wording that replaces reading | May infringe, including where a paywall or anti-scraping measure is circumvented |
What a Judgment Could Mean for Publishers
Advocate general opinions do not bind the court, although judges follow them in most cases, The Next Web noted. The Court of Justice must still deliberate. The Next Web said a final ruling is expected in the coming months and could clarify how chatbots across the EU must respect the press publishers’ right introduced in the 2019 copyright reform.
Morningstar reported that Hungarian judges will later issue a verdict in the domestic case. Google did not immediately respond to a request for comment, Dow Jones reported.
The Straits Times said a ruling that forced Google to pay fees for content use could offer a lifeline to online media whose traffic has dwindled as chatbot summaries replace search results that include hyperlinks. Publishers tracking that substitution can also read Turkey’s antitrust probe into Google AI Overviews and Reddit search referrals after Google’s AI shift.
If judges follow that split, training and live answers are separate questions for newsrooms. Authorisation, on this view, becomes relevant when retrieval bypasses a control and returns wording recognisable enough to replace the article. News Corp, owner of The Wall Street Journal, already has a commercial agreement to supply content on Google platforms, Morningstar noted. The reported opinion does not say how those supply agreements would be treated.
