🔍 Read the full analysis: The EU AI Copyright Debate, Explained For Companies, Publishers And Users on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
On Nov. 3, 2026, Advocate General Maciej Szpunar advised the EU Court of Justice that chatbots may infringe press publishers’ rights if they bypass access restrictions or reproduce an article’s distinctive expression closely enough to replace it. His nonbinding opinion says those conditions do not appear to apply to Like Company’s Gemini case; the Hungarian court will decide, and the EU court has not set a date for its judgment.
EU Court of Justice Advocate General Maciej Szpunar said on Nov. 3 that AI chatbots may infringe press publishers’ rights when they retrieve and reuse news articles in answers, but only under specific conditions. In the first generative AI case to reach the court, he said the test does not appear to be met in a Hungarian publisher’s dispute with Google over Gemini; his opinion is nonbinding, and the national court will decide the case.
The dispute began after Gemini summarized a Hungarian news article about a singer’s reported plan to bring dolphins to Lake Balaton. Like Company, which operates Hungarian news portals, sued Google in Hungary, alleging that its articles were used to train Gemini and were repeatedly reproduced through the chatbot from June 2023 to May 2024. The Budapest regional court referred four questions to the EU court: two about training and two about chatbot answers.
Szpunar’s answer on chatbot responses draws a distinction between facts and expression. Repeating the facts reported in a story is not, by itself, enough to infringe the press publishers’ right, he said. A chatbot could cross the line if it obtained an article by bypassing a paywall or an anti-scraping restriction, or if its response reproduced the article’s form of expression so recognizably that readers could avoid opening the original.
On the example before him, Szpunar said the conditions did not appear to be present: the article was not protected by the access barriers described in his test, and the sample answer did not appear to copy recognizable wording. He also pointed to safeguards Google appeared to have in place. These are assessments in an advisory opinion, not findings in a final court ruling.
Infographic 01 · Where the line runs
When a chatbot answer infringes a publisher's right
Advocate General Szpunar's test in Case C‑250/25 for chatbots that retrieve news articles in real time and reuse them in their answers.
Stays free
- Reporting the same facts. Nobody owns the news itself (pts 199–200).
- A short factual summary, presented differently (pt 208).
- An overview built from several sources, restructured, with links back (pts 208, 234).
- Individual words or very short extracts (pt 201).
Can infringe
- Getting around a paywall or anti-scraping block to fetch the article (pt 188).
- An answer that keeps the article’s structure and phrasing, even paraphrased, so readers can skip the original (pts 202, 205, 207).
- Systematically recognisable articles in answers to related questions: making available to the public (pts 213, 240).
The test, step by step
For a chatbot that fetches a news article live (RAG) and answers a user
- Is it an online use by a service in the EU?Yes for RAG answers to EU users, wherever the servers are (pts 165–169).
- Did the bot get in lawfully?No circumvention of effective technical measures (pt 188).
- Is the article’s form of expression recognisable in the answer, to the point of substitution?Side-by-side comparison; facts do not count (pts 196–209).
- Is the provider responsible?If inherent to its business model, or no reasonable measures such as respecting blocks and output filters (pts 221–222).
- Temporary copies during retrieval are then exempt under Art. 5(1) InfoSoc; the TDM exception does not apply to RAG (pts 179–189).
An Advocate General’s opinion, not a judgment. General information, not legal advice.
Sources: Opinion of AG Szpunar, C‑250/25 (provisional text), CURIA press release No 155/26 (3 Nov 2026). Point numbers refer to the provisional text.
Powered by Thorsten Meyer AI
The Line Between Summary and Substitution
The opinion could shape how companies build AI search and chatbot products that draw on current web pages. If the EU court adopts Szpunar’s approach, developers may need to account for access controls and avoid answers that reproduce a publisher’s distinctive expression closely enough to substitute for the source. The opinion does not say that every summary of a news article requires a licence.
For publishers, the proposed test offers a possible route to challenge certain forms of reuse, while leaving the underlying reporting and facts available for other services to describe. For users, the practical effects could include answers that rely more on synthesis across sources and clearer links to original reporting. Those outcomes depend on the final judgment and how courts apply it to particular systems and content.
Popular right now
Ninja Crispi Glass Air Fryer (Stone) – 4‑in‑1
Shark HydroVac MessMaster Cordless Wet Dry Vacuum and Mop Combo, AW261
Breville BES870BTR Barista Express Espresso Machine, Black Truffle
As an affiliate, we earn on qualifying purchases.
A 2019 Right Meets Chatbots
The case concerns the EU press publishers’ right under Article 15(1) of the 2019 Copyright in the Digital Single Market Directive. The right lets publishers control certain online uses of their press publications by online services. As Szpunar recalled, lawmakers had services such as news aggregators, including Google News, in view when adopting it. The dispute asks whether and how that right applies when a chatbot retrieves an article and uses it to answer a person’s question.
Like Company also alleged that Google used its articles to train Gemini. Szpunar advised the court not to answer the training questions because the case record did not establish that the articles were in the model’s training data. The dolphin story appeared in July 2023, after Gemini 1.0’s reported training period, and the user’s prompt named the site. Szpunar considered live retrieval, known as retrieval-augmented generation, a plausible explanation for the response.
He added, as a separate point, that model training may generally fall under the EU’s text-and-data-mining exception, subject to rights holders’ ability to opt out in a machine-readable form. That discussion was not necessary to decide the questions, which he considered inadmissible, and should not be read as a ruling that Google trained Gemini on Like Company’s material.
The Final Ruling Still Awaits
The Advocate General’s opinion is not binding. The Court of Justice must issue its own judgment, and no date has been announced. The Hungarian court will then apply the EU court’s interpretation to the dispute. It remains unclear whether the judges will adopt Szpunar’s proposed limits, how they will handle the specific evidence, or what remedy could follow.
The training issue also remains unresolved in this case. Szpunar did not give a definitive answer to the referred questions because he found the factual basis insufficient. His additional discussion of text and data mining is guidance in an opinion, not a ruling that resolves whether a particular model’s training complied with copyright law.
Hungarian Court Applies EU Guidance
The next major step is the Court of Justice’s judgment in Case C-250/25, after which the Budapest regional court can continue deciding Like Company’s claims against Google. The EU court has not announced a timetable. Its preliminary references took an average of 16.9 months in 2025, but that historical average does not establish when this case will be decided.
Until the judgment, companies, publishers and users should treat the opinion as a proposed legal framework rather than settled law. The central questions to watch are whether the court adopts the access-barrier and recognizable-expression tests, and how it distinguishes live retrieval and chatbot answers from model training.
Key Questions
Has the EU Court of Justice ruled that Gemini infringed copyright?
No. Szpunar issued a nonbinding opinion on Nov. 3, 2026. The court has yet to issue its judgment, and the Hungarian court will decide the underlying case.
When could a chatbot’s news summary infringe publishers’ rights?
Under the Advocate General’s proposed test, infringement may arise if the chatbot obtains an article by bypassing an access restriction or reproduces its expression so recognizably that readers can avoid the original. Reporting the same facts alone is not enough under that approach.
Did the opinion establish that Google trained Gemini on the publisher’s articles?
No. Szpunar said the case record did not show that Like Company’s articles were used in training and advised the court not to answer those questions. He considered live retrieval from the website a plausible explanation for the example summary.
Does the opinion mean AI companies can train on any copyrighted work?
No. The opinion’s discussion of the EU text-and-data-mining exception was additional commentary, not a ruling on the use of any specific work. It also refers to rights holders’ ability to opt out in a machine-readable form.
Source: ThorstenMeyerAI.com
