Skip to content
All articles

Peri News / European Union

EU court adviser backs copyright exception for AI training, with limits

According to PPC Land, the Advocate General considers the text and data mining exception applicable to AI training under certain conditions. His opinion is non-binding, and the Court may leave the issue unanswered.

An open book and paper strips passing through an opening in a blue panel toward a chip, a metaphor for using texts to train AI within limits.
AI-generated illustration

According to PPC Land’s November 5, 2026 article, “EU court adviser in Google case backs AI training on 3 conditions”, Advocate General Maciej Szpunar considers that AI training on protected books, articles and other works can, in principle, benefit from the EU text and data mining exception. His opinion, delivered on November 3 in case C-250/25, Like Company v Google Ireland, is non-binding. Szpunar first recommends declaring the training questions inadmissible: there would be insufficient evidence that the disputed articles were used to train Gemini, and the right invoked by the publisher would not cover that activity. The Court could therefore leave the issue unanswered.

In PPC Land’s account, the provisional analysis identifies three conditions for relying on the exception under Article 4 of the Copyright in the Digital Single Market Directive. Works must be lawfully accessible: developers should avoid obviously unlawful sources and must not circumvent paywalls or effective measures against automated access. They must also respect rights reservations expressed in an appropriate manner, including through robots.txt for online content. Finally, applying the test that limits copyright exceptions to safeguard exploitation of works, Szpunar would require developers to demonstrate measures against excessively close reproduction of specific training material, such as dataset deduplication and filters. Without that evidence, the national court should refuse the exception.

PPC Land further reports that Szpunar distinguishes the press publishers’ related right over press publications, established by Article 15 and limited to online uses, from copyright in individual articles: the former would not cover training, while the latter would remain relevant. The text and data mining exception would also not cover model outputs, whose possible infringement would require assessment case by case. The territorial reach of EU law where training occurs abroad remains unresolved. The opinion itself has not been consulted directly; this account relies on the journalistic report.

Sources

Privacy and cookie notice ·