The very recent judgment handed down by the Regional Court of Munich in the GEMA v. Suno case (Regional Court of Munich, First Division, decision of 31 July 2026, Case No. 42 O 763/25) represents one of the most important European developments in the debate on the relationship between generative artificial intelligence and copyright. The decision, which is still subject to appeal, addresses the main legal issues relating to the training of AI models based on third-party copyrighted works, the memorisation of content, and the application of European law also to platforms operating on a global scale.
GEMA, the German collecting society (the equivalent of SIAE in Italy), challenged Suno – a platform that enables users to generate musical works from text prompts – over its use of works in GEMA’s repertoire for training purposes and for generating songs that were very similar to well-known tracks (including hits such as “Mambo No. 5” and “Daddy Cool”), arguing that such activities constituted acts of reproduction and memorisation of copyrighted works without authorisation.
The judgment found that Suno’s model had in fact been trained using works from GEMA’s repertoire and that it operated by retaining copies necessary to generate the content requested by users, characterising such conduct as copyright infringement. The Court ordered Suno to cease such uses and to disclose documentation relating to the platform’s use of the works and its revenues for the purposes of assessing damages.
Of particular interest is the Court’s territorial approach: the fact that the training had taken place in the United States did not exclude the unlawful nature of the conduct, given that its effects occurred in the EU through a platform accessible to European users. The Court excluded the exculpatory effect of the so-called “fair use doctrine” under US law (even if it were considered applicable), as well as that of the European text-and-data-mining (TDM) exception, particularly on the ground that the scraping of content (which Suno had apparently carried out on platforms such as YouTube) had resulted in the acquisition of the works in their entirety, rather than merely the extraction of data relating to them.
The German judges also addressed the issue of the concept of a “copy” in the context of generative AI. According to the Court, the storage of a work constitutes an unlawful copy even where the work is not displayed “in plain view” to the user, provided that its entire content can nevertheless be retrieved through a simple open-ended prompt. This legal interpretation of the concept of a copy in the new context of generative AI is of particular interest because, for example, it differs from the approach taken in November 2025 by the High Court of England and Wales in another high-profile case, Getty Images v. Stability AI, concerning an image-generation platform (High Court of England and Wales, [2025] EWHC 2863 (Ch)).
The German decision under discussion is also particularly significant because it applies these principles for the first time to the music sector, where the assessment of similarities between works is traditionally more sophisticated than in other creative fields. From this perspective, the judgment is of greater interest than the precedent obtained by GEMA in Germany in 2025 against OpenAI concerning the reproduction of song lyrics alone.
Suno has already announced its intention to appeal the decision. Nevertheless, this is a precedent that provides extremely relevant guidance on the application of the EU legal framework to AI and model training, adding an important layer of protection from an intellectual property perspective, beyond those already implemented at the regulatory level through the transparency obligations introduced by the AI Act. The European approach increasingly appears to be aimed at ensuring adequate remuneration for rights holders, while encouraging models based on fair and transparent licenses, terms and conditions.
It is certainly significant that, immediately following the publication of this decision, a sort of international “chain reaction” has already been triggered over the past few weeks. On the one hand, major companies such as BMG and Warner have announced a new licence with Suno aimed at “regularising” the platform’s future use of copyrighted content; on the other hand, Sony promptly brought proceedings against Anthropic over the platform’s use of copyrighted content in connection with Claude. Meanwhile, the US administration filed, in another set of proceedings pending against OpenAI before the Southern District of New York, a Statement of Interest in which it strongly argues that the use of third-party works to train GenAI platforms does indeed constitute “fair use” (contrary to the position taken in the GEMA judgment discussed here) and should therefore, in its view, be broadly tolerated rather than sanctioned.
This intensification of the debate and litigation is likely to have significant implications in Italy as well, where a growing number of similar claims and disputes are likewise expected, such as the proceedings, still pending, brought late last year by RTI and Medusa Film before the Court of Rome to protect their original audiovisual content against the Perplexity AI platform. More generally, rights holders and owners of protected content (as well as collecting societies) are becoming increasingly active worldwide in an effort to prevent not only works, but also performances, artistic interpretations and distinctive identifying elements, from being unlawfully and freely exploited by these platforms.
