EU adviser says Gemini answers may infringe publishers' rights
On 3 November 2026, Advocate General Maciej Szpunar of the Court of Justice of the European Union said that Google's Gemini chatbot may infringe press publishers' rights when it retrieves a news article, reuses it to write an answer, and shows that answer to a user. The opinion arises from a case brought by Like Company, a Hungarian publisher, in Case C-250/25. The opinion is not binding, and the Court is expected to rule some months later. In the same opinion, Szpunar said EU text and data mining rules generally allow AI training on copyrighted material, and he found no evidence that Gemini was trained on the publisher's articles. For businesses that automate work with AI, the practical point is what a tool retrieves and what it writes back.
In short
- On 3 November 2026 an EU Advocate General said Gemini answers that reuse press articles may infringe publishers' rights.
- The opinion is non-binding and the Court of Justice has not yet ruled.
- Training on copyrighted material was described as generally allowed under the EU text and data mining exception.
- Businesses should check what their AI tools retrieve and what they write back to customers.
- Review of flagged drafts stays with a person, and legal disputes need a lawyer.
What happened
On 3 November 2026, Advocate General Maciej Szpunar delivered an opinion at the Court of Justice of the European Union in Case C-250/25, Like Company v Google Ireland. The Court's own press release describes it as the first generative AI case to reach the Court. The referring court sits in Budapest, and the claimant is a Hungarian publisher that says Google used its articles through Bard, now Gemini.
Szpunar's central point concerns answers. When an LLM-driven chatbot retrieves a publication from a press publisher's website, reuses it to generate an answer, and gives that answer to a user, the press publisher's right may be infringed in certain circumstances. He stressed that the right covers the specific form of expression in a publication, not the facts or the news itself. According to the Court's summary, two situations carry the most risk: access gained by getting past paywalls or other technical barriers, and answers that reuse the publication's expression in a recognisable way.
Training is treated differently
On training, the opinion is more favourable to AI developers. Szpunar said that training large language models on copyrighted material is freely permitted under the text and data mining exception in EU law. He also found no evidence that Gemini had been trained on the Hungarian publisher's articles, and he advised the Court to treat the training questions as hypothetical on that record.
That exception rests on Article 4 of Directive (EU) 2019/790. It covers reproductions made for text and data mining of lawfully accessible works, but rights holders can reserve their works. For content published online, the reservation has to be made in a machine-readable way, such as through metadata or website terms.
Why it matters for businesses that automate work
Most business automation does not train public models. It does, however, often feed contracts, emails, supplier portals and web pages into a tool that drafts replies, reports or summaries. The opinion draws its line between what a tool learns and what it shows to a user. A draft that copies recognisable wording from a third-party article is the part the adviser flagged. Any business that lets an AI assistant place third-party text into customer messages or client documents now has a copyright question to answer, not only a quality question.
Google's response, as reported by Reuters through secondary coverage, stressed that Gemini learns linguistic patterns to create original outputs rather than replicating content. The opinion is advice to the Court, so it does not settle that argument for any business.
What changes in practice
The opinion does not create new duties for a business that uses AI to automate its own paperwork. It does show where the exposure sits. The table below sets out the checks that map to the adviser's reasoning.
| Point in the workflow | What to check | Why it matters under the opinion |
|---|---|---|
| Sources the tool reads | Whether the material is lawfully accessible and whether any access control was bypassed | Lawful access and technical barriers are central to the adviser's test |
| Drafted output | Whether it repeats recognisable passages from a third-party article or report | Protection covers the specific expression, not the underlying facts |
| AI vendor terms | What the vendor says about training and about retaining your inputs | Training and output are assessed differently in the opinion |
| Your own published material | Whether your site or documents carry machine-readable reservations | Article 4 relies on reservations for online content |
Automation cannot judge copyright on its own. A person still decides whether a flagged draft copies protected wording, and a lawyer is needed for any actual dispute. Smaller firms without in-house legal staff should set the review rule in advance rather than deciding case by case.
What to watch next
- The Court of Justice judgment in C-250/25, which normally follows some months after an opinion.
- Whether the Court accepts the adviser's split between answers and training, since the Court is not bound by the opinion.
- The Budapest court's application of the final ruling to the Gemini answers in the case.
- Two further pending Gemini referrals, which a Hungarian law professor cited in press coverage as the next steps.
This explains the rule in general terms; it is not legal or tax advice for your case.
Sources
How AiStaffo would automate this
AiStaffo would connect the document inbox, customer message tool and report templates a business already uses, so that each AI-drafted reply is logged with the source files it drew on. The system would run the routine steps automatically: sorting incoming material by source, flagging any draft that repeats long passages from a third-party file, and keeping a record of what was retrieved. The owner or a named staff member still reviews flagged drafts, decides which sources the business may use, and approves anything sent to customers. Book a free automation audit.
Questions people ask
Has the EU court ruled that Google's Gemini infringes copyright?
Does the opinion say AI training on news articles is illegal in the EU?
What exactly may infringe the press publishers' right?
Does this affect a business that only uses AI for internal paperwork?
When will the Court of Justice decide the case?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.
