Update on the article “Do you hear the same?“: On 9 March, GEMA and the US company Suno Inc. had their first exchange of blows at Munich Regional Court I on the question of the extent to which AI-generated music infringes the rights of authors of protected musical works.
Background
In contrast to the much-noticed AI decision last year, in which GEMA – before the same court – won a first-instance victory against OpenAI (see LG Munich I, final judgement of 11 November 2025 – 42 O 14139/24), the current proceedings are not about song lyrics, but about pieces of music as such.
Suno enables users to create complete songs via its “Music AI Generator” by simply entering lyrics (so-called “prompting”). The AI tool not only provides lyrics, but also a melody and vocals on request. Even mood, style and genre can be specified.
GEMA has also tested this out and used the Suno AI to produce AI output for the six original songs “Forever Young”, “Atemlos”, “Mambo No. 5”, “Rasputin”, “Big in Japan” and “Daddy Cool”. In each case, she prompted the original lyrics of the song, the desired musical style and the title of the work. However, the melody, rhythm, harmony or arrangement were not specified.
As the resulting output delivered by the AI was very similar to the original songs, GEMA considered the artists’ copyrights to have been infringed in several respects. The collecting society therefore filed a lawsuit against Suno last year at the regional court for injunctive relief, information and damages.
AI training with original songs
One thing is clear: Suno has trained its music generator with the six original songs mentioned above, which GEMA has made the subject of its lawsuit. At least the parties agree on this.
In order to extract the musical works from the YouTube platform, Suno used so-called stream ripping techniques. In this way, technical protection measures implemented by YouTube (so-called “rolling ciphers”) are circumvented, which are intended to prevent the downloading of audio and video content.
Collecting society vs. AI provider – the arguments
GEMA sees the AI-generated outputs of the music pieces as proof that the works have been memorised in the AI. The almost identical sounding pieces would constitute an unauthorised reproduction of the copyrighted original work. The output in itself is also an infringement of rights because it is a case of making available to the public. GEMA considers the rights of the authors to be infringed both in Germany and in the USA.
Suno, on the other hand, rejects any allegations. The company already believes that the German court seised does not have jurisdiction. In addition, its actions are covered by “fair use”, a US legal institution according to which unauthorised use of protected material is permitted for certain purposes (e.g. for academic purposes or satire).
According to Suno, there is also no recognisable infringement under German copyright law. Firstly, in the opinion of the AI company, the songs in dispute are not recognisable in the outputs. Moreover, the AI only depicts mathematical patterns as well as syntactic, semantic and contextual relationships that it has learnt from the training data. However, the training data itself is neither contained in the AI model nor stored in it. The outputs at issue were also attributable to the refined prompts of the respective users and the respective output was therefore attributable to them. Suno also brought the text and data mining barrier into play, which could be used to justify any potential infringements.
Parallels to the GEMA-OpenAI case
In any case, Open AI did not get any further with these arguments before the Munich I Regional Court last year. With regard to AI language models, the court considered it proven at the time that OpenAI reproduces the protected content in the original wording and makes it accessible to the user, i.e. “memorises” it.
With regard to AI training and memorisation, the court clearly saw the AI company as being responsible for the output, not the prompt-generating user.
With regard to text and data mining, the court made it clear: AI training only falls under this exception if it involves the automated extraction of information patterns and mere acts of reproduction or analysis are carried out without economic harm to copyright interests. However, if entire works can be reproduced – i.e. “memorised” – from the model, this no longer has anything to do with mere analysis.
If the court now applies the same standards in the Suno case, there is much to suggest that a decision will also be made in favour of the rights holders here.
Outlook – What happens next
Last Monday, the court first discussed the factual and legal situation of the case. Suno now has until 7 April 2026 to respond in writing to GEMA’s allegations. The chamber has also set a hearing date. A decision in the case is scheduled for 12 June 2026.
So it remains exciting. We will be happy to keep you up to date.