EU Court Adviser Says Gemini Answers Built From News Articles May Infringe Publisher Rights
An Advocate General at the EU's top court says Google's Gemini may infringe publishers' rights when it reuses news articles in answers, while AI training on public text is generally allowed. His opinion is not binding and the court has not ruled yet.
Google got a rough headline on Tuesday and a more mixed legal reality. An adviser to the EU's top court says Gemini can infringe publishers' rights when it grabs a news article and rebuilds it as an answer. The same adviser says training AI on public text is, as a rule, allowed.
The opinion comes from Advocate General Maciej Szpunar in Case C-250/25, Like Company, the first generative AI case to reach the Court of Justice of the European Union, according to the court's press release. An Advocate General's opinion is not binding. The judges are only beginning their deliberations, and judgment will come at a later date. Reuters reports the court follows the majority of its advisers' opinions and will rule in the coming months.
Like Company, a Hungarian operator of web portals, told the Budapest Regional Court that Google used its articles to train Gemini and then reproduced them in chatbot answers between 13 June 2023 and 31 May 2024. The Hungarian court sent four questions to Luxembourg, two on training and two on answers.
Szpunar wants the court to skip the training questions. He calls them hypothetical, because there is no evidence Like Company's articles trained Gemini, and because training is not an online use of press publications under the publishers' right in question. He adds, for completeness, that training a large language model on protected works is as a rule freely permitted under the text and data mining exception.
"When an LLM-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." Advocate General Maciej Szpunar
Those circumstances are narrower than the alarm suggests. Per the press release, the right covers a publication's specific form of expression, not the facts it reports, and exists to protect investment, not to give publishers a monopoly over the news. Reproduction turns on two factors. Did the chatbot get in unlawfully, say by circumventing a paywall? Or does the answer reuse the article's form of expression so recognisably that the reader is entirely dispensed from visiting the original? The provider is liable where such uses are inherent to its business model, or where it has not put credible and effective countermeasures in place.
Read the fine print before naming a loser
Here is what the setback framing glosses over. João Pedro Quintais of the Institute for Information Law, writing on the Kluwer Copyright Blog, notes that when Szpunar applies his framework provisionally to the Hungarian example, the conditions for liability do not appear to be met. The article's form of expression did not seem recognisable in the response, and Google appeared to have measures to mitigate the risk.
Google sees it that way. A spokesperson told Reuters that text and data mining exceptions are "vital for Europe's competitiveness and AI development," and welcomed the recognition that EU law permits AI training on publicly available information. Gemini, the spokesperson said, "learns linguistic patterns to create original outputs rather than replicating content," with publisher controls that let rightsholders opt out.
Publishers heard something else. Angela Mills Wade, executive director of the European Publishers Council, wants search indexing kept apart from authorising AI answers that substitute for original reporting.
"Publishers should not have to choose between search visibility and protecting their rights." Angela Mills Wade, European Publishers Council
Reuters says the case lands amid growing friction over chatbots using press stories without what publishers call proper compensation, and notes that online media have seen traffic dwindle as AI summaries replace search results with links.
My read: the real fight is not scraping in the abstract. It is the word recognisable, and who decides when a paraphrase is close enough to replace the story. Expect it argued article by article. Training, the issue AI labs fear most, got a friendlier signal, though the adviser asked the court to leave it unanswered. Our earlier piece on the AI training opt-out is worth a reread.
The judges can adopt Szpunar's framework, trim it or toss it. Until they rule, nobody knows what Gemini owes publishers, if anything.
