German court’s AI ruling ‘remarkable’
Lawyers at Pinsent Masons say that a German regional court has, for the second time within nine months, handed down what they describe as “a fairly remarkable” judgment in an AI-related copyright matter.
They highlight that, in the latest case, the court made findings against an AI developer not only under German copyright law, but also under US copyright law.
In an analysis on the firm’s website, the lawyers note that, in both cases, the judges considered claims brought by the German collective rights society GEMA, which mainly focuses on administering and licensing music copyrights on behalf of composers, lyricists, and music publishers.
GEMA successfully sued OpenAI for copyright infringement in a judgment handed down in November 2025. Its latest success came against US-based Suno, which operates an AI-based music generator.
Data from YouTube
Suno’s tool was trained in the US with data derived from YouTube, among other sources.
The court determined that the training data was obtained by illegal means, such as stream-ripping, which involves circumventing security measures applied by online streaming service providers in order to download the content – in this case for third-party AI training purposes.
Pinsent Masons notes that the judges in Munich assumed jurisdiction not only over the provision and use of the Suno music generator in Germany, but also in respect to the training of the underlying AI model that took place on US soil.
Accordingly, they applied German copyright law to the operation of the AI-based application and US copyright law to the training phase.
The Munich court upheld the asserted claims, assessing that Suno required a licence from GEMA to use the works in question to train its AI model, and to offer the creation of music pieces by means of Suno’s music generator.
‘Memorisation’
“As regards the latter, the judges took the view – as they had done in the OpenAI case too – that copyright-protected works which had been used in the course of the training phase would continue to exist within the AI model by means of systematic memorisation,” the Pinsent Masons lawyers state.
They add that Suno, rather than the prompting user, was seen by the judges as the main creator of the infringing output, given that fairly simple prompts repeatedly led to output being generated that was either identical or at least very similar to the protected lyrics and melodies.
In reaching its verdict, the court rejected Suno’s arguments that its system had neither stored nor reproduced the copyright works at issue, pointing to the phenomenon of ‘memorisation’.
The Pinsent Masons lawyers point out that the concept of memorisation is “heavily debated” among copyright experts, with some disputing that AI systems memorise the content they are trained on.
‘Fair use’ principle
They add that the Regional Court of Munich took a different perspective and ruled in favour of GEMA, after closely examining the requirements of the ‘fair use’ principle – a carve out from copyright protections provided for under US copyright law.
“Ultimately, they reached the conclusion that Suno’s use of the respective works went beyond what may be justified under the doctrine of fair use,” they lawyers say.
“The court further considered that the reproduction of the copyright-protected music by Suno did not fall within the text and data-mining exception contained in German, and EU, copyright law either,” they add.
GEMA has raised separate litigation against Suno in respect of its AI training before the US courts.
Pinsent Masons points out that, earlier this year, the US government said that it believed that AI developers could use others’ copyrighted content to train their models without infringing US copyright laws.