Artificial intelligence may learn from news articles. However, if a chatbot reproduces them so faithfully in its responses that readers no longer have any reason to open the original text, it may infringe publishers’ rights. This was the conclusion reached on Tuesday by Advocate General of the Court of Justice of the European Union Maciej Szpunar in a dispute between a Hungarian media company and Google over the Gemini chatbot. This is the first time the Luxembourg court has addressed generative artificial intelligence.
Hungarian publisher takes on Google
Like Company operates several news websites in Hungary. It sued Google’s Irish subsidiary because, according to the company, Gemini generated summaries of its articles in response to users’ queries. The publisher claims that Google used its texts to train language models and repeatedly copied and made them available to the public through the chatbot without permission. According to the lawsuit, these were not merely short excerpts tolerated under European rules. The dispute concerns the period from mid-June 2023 to the end of May 2024.
In April 2025, a Budapest court asked the Court of Justice of the EU to interpret EU rules. The case, designated C-250/25, became the first in which European judges are examining the relationship between copyright and generative artificial intelligence.
Google has defended itself from the outset. It claims that Gemini does not store articles in its memory and that any disputed responses fall under the exceptions for temporary copies and text and data mining. The company also challenged whether Hungarian law applies to the dispute at all, because Gemini was not trained in Hungary.
Training yes, copying no
In his opinion, Szpunar distinguished between two issues that are often conflated in debates about artificial intelligence. The first is the training of the model itself. EU law freely permits the use of protected texts to train language models precisely because of the text and data mining exception. The Advocate General also noted that there was no indication that Gemini had been trained directly on Like Company’s articles. He therefore recommended that the Court declare two questions concerning training inadmissible and focus on the responses the chatbot gives users.
And this is where the problem begins. Gemini does not rely solely on what it learned during training. It can also search the internet in real time and base its responses on the information it finds. In doing so, it may also retrieve newly published news articles. When a chatbot downloads text from a publisher’s website, uses it to compose a response and then provides that response to a user, it may, under certain circumstances, infringe the publisher’s rights.
However, Szpunar did not give publishers a blank check. Copyright does not protect the facts reported in articles themselves, but only the specific form in which the editorial team presented them. No media outlet can therefore own the news that an event occurred. It can only protect the way it wrote about it. According to the Advocate General, the purpose of the rules is to protect the money and work invested in producing an article, not to give publishers a monopoly over news.
When a response replaces the entire article
To help courts assess whether a chatbot has crossed the line, the Advocate General offered two guidelines. The first concerns how it accessed the article in the first place. A problem arises primarily when it circumvents technical protections, such as paid access or barriers against automated content scraping.
The second guideline concerns the response itself. What matters is whether the chatbot merely conveys facts or whether readers can recognize the specific treatment used in the original article in its text. It is also important whether the reproduction goes so far that users no longer need to open the original publication. According to Szpunar, if a chatbot consistently reproduces the form of articles to such a significant extent in responses to similar queries, it may also infringe the publisher’s right to make content available to the public.
The company operating the chatbot may then bear direct liability. This will be the case if such use of articles is an integral part of its business model, or if it has failed to introduce measures that would credibly and effectively prevent it.
Google responded to the opinion in a conciliatory manner. Its spokesperson said that Gemini learns language patterns and creates its own original texts instead of copying content. Publishers have access to a range of tools allowing them to decide whether they want to opt out of having their content used. The company also welcomed the Advocate General’s recognition that training artificial intelligence on publicly available information is legal.
The Advocate General’s opinion is not binding on the Court, but the judges follow it in most cases, and it is regarded as a strong indication of the direction the case is taking. A judgment is expected in about six months. It will show how chatbots throughout the European Union must respect publishers’ rights introduced into EU rules by the 2019 copyright reform.
Publishers and technology companies are also fighting over the use of journalistic work on the other side of the Atlantic. The New York Times and three other US newspapers recently sued Brave over the scraping of content for artificial intelligence. But the Luxembourg dispute is focused elsewhere. It is concerned less with what models learn from than with what they actually present to people afterward.
Sources: reuters.com and thenextweb.com
