EU court adviser says Gemini answers can infringe press rights, training mostly lawful
An EU Advocate General said on 3 November 2026 that Gemini answers built from news can infringe press rights, while AI training is mostly lawful.
By Gürkan E.Published
Tech AI LatestAdvocate General Maciej Szpunar delivered his opinion in Like Company v Google Ireland (C-250/25) on 3 November 2026, the first generative AI case before the EU Court of Justice. He said training language models is, as a rule, covered by the text and data mining exception, but that retrieving press articles to answer users can infringe publishers' rights in two situations. The opinion is not binding; the Court rules later.
Key facts
- Delivered: 3 November 2026 by Advocate General Maciej Szpunar (opinion in Case C-250/25)
- Parties: Like Company, a Hungarian web portal operator, against Google Ireland over Bard and Gemini answers
- Training: as a rule covered by the text and data mining exception; the training questions should be declared inadmissible
- Answers: retrieving press content to generate a reply can be an online use under Article 15 of Directive 2019/790
- Status: advisory only; the Court of Justice's judgment is expected later, date not stated
What the Advocate General said
Advocate General Maciej Szpunar gave his opinion in Case C-250/25, Like Company v Google Ireland, on 3 November 2026. It is the first case about generative AI to reach the Court of Justice of the European Union, and the first time a member of the Court has set out how EU copyright law applies to a chatbot. The opinion advises the judges; it does not decide the case.
Like Company runs Hungarian news portals. It sued Google in a court in Budapest under the press publishers' right in Article 15(1) of the EU Copyright Directive, Directive (EU) 2019/790. It says Google infringed that right between 13 June 2023 and 31 May 2024, first through its Bard chatbot and then through Gemini, by answering users with content taken from its articles. The Hungarian court asked the Court of Justice how the directive applies to the way a chatbot is trained and to the answers it gives.
On training, Szpunar advised the Court to treat the questions as inadmissible. According to the opinion as summarised by IPWatchdog, there is no evidence that Like Company's articles were used to train the Gemini models, and training is not an online use covered by Article 15 in any case. He added that training a large language model is, as a rule, covered by the text and data mining exception in Article 4 of the directive.
On answers, he reached a different view. Retrieving press publications in real time to generate a chatbot response can, he said, amount to an online use under Article 15. He described two situations in which that use infringes the publisher's right: when the chatbot reaches the content by getting around a paywall or another technical barrier, and when the answer reproduces the publication's expression so recognisably that the user no longer needs to read the original. The provider is liable when that use is inherent to its business model.
The limits he placed on the publisher's claim
The opinion does not hand Like Company a win. Szpunar noted that the article at the centre of the dispute appears to be a purely informative report of events, and he raised doubts about whether portals like Like Company's qualify as press publishers at all, without resolving the point. He also observed that the publisher apparently had not installed effective measures to block scraping.
The text of the opinion records that Google and the European Commission compared the disputed article with a Gemini answer and found that the two contained essentially the same facts in different wording. Facts are not protected by copyright, so on that example the answer did not clearly reproduce the article's expression. Szpunar also suggested that the copies made while a chatbot retrieves and processes a page may fall under the temporary reproduction exception in Article 5(1) of the InfoSoc Directive.
What changed and why it matters
Until now, EU courts had not said how the press publishers' right, created in 2019 mainly with news aggregators and search snippets in mind, applies to AI assistants. The opinion separates two activities that publishers often treat as one. Training, the step most lawsuits in the United States focus on, would mostly fall under the existing text and data mining exception, which publishers can opt out of with machine-readable reservations. Answering users with retrieved news, the step that competes most directly with a visit to the publisher's site, could need a licence where it bypasses a paywall or copies the expression of the article.
That split matters for every company running retrieval-based assistants in Europe, not only Google. It points publishers towards technical access controls and opt-outs, and it points AI providers towards answers that summarise facts rather than reproduce text. It also fits a wider pattern: the European Commission is consulting on copyright and AI until 10 November 2026, major developers have agreed on a robots.txt signal for AI training, and in the United States the Justice Department has backed OpenAI's fair use position in the publishers' cases. In March 2024 the French competition authority fined Google 250 million euros in a dispute that included the use of publisher content for its Bard model.
What to watch next
More on EU and US rules for AI is in our AI policy coverage. The Court of Justice now deliberates and will deliver its judgment later; no date has been set. The judges follow an Advocate General's opinion in many cases but not in all, and they may answer the referred questions more narrowly. After the judgment, the Budapest court will apply the answers to the facts, including whether Like Company is a press publisher and whether Gemini's answers reproduced protected expression.
Open questions remain. The opinion, as reported, does not settle how a provider should prove that a given article was not used in training, and it does not set a threshold for how close a summary must be before it infringes. Publishers and AI providers will also read it against the Commission's review of the 2019 directive, which can start from June 2026 under Article 30 of the directive.
What changed
| Before | No EU court had addressed how the press publishers' right and the text and data mining exception apply to generative AI chatbots |
|---|---|
| After | The Advocate General advises that training is mostly covered by the exception, while real-time retrieval of news for answers can infringe the press publishers' right in two situations |
| Who it affects | Google and other providers of AI assistants in the EU, European news publishers, and national courts hearing AI copyright claims |
Questions people ask
- Is the Advocate General's opinion in Like Company v Google binding?
- No. It advises the Court of Justice, which will rule later. The judges often follow the Advocate General, but they are free not to.
- Does the opinion say AI training on news articles is legal in the EU?
- Szpunar said training a large language model is, as a rule, covered by the text and data mining exception in Article 4 of Directive 2019/790, but he advised the Court not to rule on training in this case because there is no evidence Like Company's articles were used.
- When can a chatbot answer infringe a press publisher's right?
- According to the opinion, when the chatbot reached the article by getting around a paywall or other technical barrier, or when its answer reproduces the article's expression so closely that users need not read the original.
Sources
- Opinion of Advocate General Szpunar in Case C-250/25, Like Company v Google Ireland (Court of Justice of the European Union, 3 Nov 2026)
- CJEU Advocate General Outlines Copyright Limits for AI News Summaries (IPWatchdog, 4 Nov 2026)
- Setback for Google as EU court adviser warns AI chatbot may breach copyright rules (TechCentral.ie, 4 Nov 2026)
Reference pages
- Google company
- EU Copyright Directive (Directive (EU) 2019/790) policy
Spotted an error? See the corrections policy and tell the editor. Facts in this story come from the sources listed above; vendor claims are attributed.
Comments
Loading comments.
