ShineWhere the World Can See You

Generic filters

Gemini’s EU Copyright Test Separates Training From the Answer It Gives

Europe’s first generative-AI case at the Court of Justice has produced an important distinction for creators: the legal analysis of model training may not be the same as the legal analysis of what an AI system retrieves and delivers in a live answer.

On November 3, Advocate General Maciej Szpunar issued a non-binding opinion in Case C-250/25, Like Company v Google Ireland. The dispute concerns Google Gemini and articles published by a Hungarian media company. The opinion does not decide the case, but it gives the Court a proposed framework for examining retrieval-augmented generation—the process through which a chatbot reaches beyond its trained model to collect current information from the web.

The answer, not only the training set

Much of the public debate around generative AI has focused on training: what was copied, whether permission was obtained, and whether text-and-data-mining exceptions apply. The Advocate General advised the Court that the two training questions in this particular case were hypothetical because there was no evidence that the publisher’s articles had been used to train Gemini.

He nevertheless suggested that training on protected material is, as a rule, permitted under the European Union’s text-and-data-mining exception. That part of the opinion will be controversial among rights holders, but it is not the only—and perhaps not the most immediately consequential—part of the analysis.

The opinion asks the Court to concentrate on the chatbot’s real-time conduct. Gemini uses retrieval-augmented generation to reach online material when answering current questions. According to the proposed framework, publisher rights may be infringed when the system unlawfully accesses a publication, including by circumventing effective technical protections, and then reproduces its protected expression recognizably enough that the user no longer needs to visit the source.

Facts remain different from expression

The Advocate General stressed that copyright-related publisher rights do not create ownership over reported facts. What they protect is the particular form in which those facts are expressed and the investment embodied in the publication. That distinction matters well beyond journalism.

For visual artists, a parallel question appears whenever an AI product retrieves a current portfolio, exhibition image, catalogue essay or interview and turns it into an answer, a description or a new visual output. The existence of factual information about an artwork does not automatically authorize copying the photograph, reproducing the artist’s language or substituting for the licensed publication in which that material appeared.

The opinion also proposes that a provider may be responsible when the disputed use is inherent in its business model or when it fails to implement credible and effective safeguards. That moves the inquiry away from the idea that every output is solely the user’s act. Product design, access controls, source handling and output behavior can all become relevant.

What artists should document now

The eventual judgment may revise, narrow or reject the Advocate General’s approach. Artists should therefore avoid treating this opinion as settled law. It is, however, a useful signal about the kinds of evidence courts may examine.

When an AI answer appears to reproduce protected material, preserve the query, the complete output, the date, the model and product version, the cited links and the original source. Record whether the source was public, licensed, restricted or behind a technical barrier. Compare the output with the protected work itself, not merely with the underlying facts. A recognizable sequence, distinctive description, composition or other expressive choice will matter more than a shared subject.

Creators who publish through websites should also keep dated records of access rules, crawler preferences, licensing terms and any notices sent to a platform. Those measures do not guarantee that a court will find infringement, but they make the history of permission and access more legible.

A legal line still being drawn

The Court’s judges are now deliberating, and a judgment will follow later. The national court in Hungary will ultimately resolve the underlying dispute in accordance with the Court of Justice’s interpretation.

For artists, the deeper lesson is that “AI copyright” is not one question. Training, retrieval, reproduction, attribution, access and market substitution are separate acts that may carry different legal consequences. A system can be permitted to learn from large bodies of material and still face scrutiny for what it fetches, how it transforms that material and whether its answer replaces the original creative source.

ATIM approaches artificial intelligence with curiosity, independence, and balance. We explore its extraordinary potential to empower artists while examining the questions it raises around authorship, copyright, consent, creative identity, transparency, and human expression.

We are neither advocates for AI nor opponents of it. Our focus is how the technology is used—and what that means for artists.

Above all, our commitment is to artists. We look at every development through the lens of what can support, protect, and empower the creative community—identifying opportunities, examining potential risks, and providing artists with the information they need to make informed decisions about their work, their rights, and their future.

Spread the Love!

ART + INTELLIGENCE

Introducing ATIM AI

Meet Your AI Creative Guide

by Viviana

We put Viviana's years of guiding artists into an AI made for you — ask it anything, day or night.

Ask it about

how to price and sell your work · protect your rights · approach galleries · grow your audience · understand how AI affects your art or how to stay inspired!

ArtTour International
0
Would love your thoughts, please comment.x
()
x
ArtTour International
Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.