The GEMA – the German music‑rights society – sued OpenAI, claiming that its chatbot ChatGPT had been trained on copyrighted song lyrics without permission. On November 11th, 2025, the Munich I Regional Court ruled in favour of GEMA, finding that both the memorization of lyrics within the AI model and their reproduction via ChatGPT’s outputs constitute copyright infringement, and ordered OpenAI to pay damages and obtain proper licences. The ruling sets a landmark European precedent: generative‑AI providers cannot assume that training on protected works is legal under text‑and‑data‑mining (TDM) exceptions, and must secure licences to use copyrighted material.
For tech startups in Germany, the impact of this ruling could be substantial. If the appeal does not overturn it, this decision confirms that companies must secure proper licensing when using third-party data for training purposes. The decision sheds light on how courts interpret “memorization” in AI, stating that companies developing GenAI can be held liable if their outputs reproduce copyrighted materials.
Tech startups building GenAI are the first to be targeted by this ruling. Although it does not directly target companies using off-the-shelf GenAI, startups relying on pre-existing models should stay alert, as arguments from this case can be extrapolated to their situation. If their products regurgitate copyrighted works without proper licensing, these startups might also be infringing protected content.
Publicly available information is not automatically free to use. The challenge for tech startups lies in the opacity of pre-existing GenAI models, both technically and legally, since a startup is often unable to fully assess whether those models were lawfully trained and to what extent their own products might be contaminated when building on top of them.
An appeal is underway, and this court ruling is only a national one, meaning there are no legally binding effects that automatically extend to markets outside Germany. A startup in France or Sweden is not legally bound by this ruling.
However, this decision should be on startups’ radar across the EU, mainly because the court’s arguments heavily rely on the interpretation of an exception covered by an EU Directive on copyright law. There is a chance other EU states will “borrow” this interpretation, since copyright laws are largely harmonized throughout the EU. Judges across the EU remain free to interpret similar cases at their discretion until an ECJ ruling provides clarity.
This German ruling stands in sharp contrast to recent US and UK decisions, which could generally allow AI training under fair use or find no infringement if outputs do not copy protected works. Such divergence creates uncertainty for startups operating globally, meaning that what is safe in the US or UK may still trigger liability in the EU, making jurisdiction-specific compliance harder to accomplish.
While the GEMA v. OpenAI ruling raises important legal questions, it should not stifle innovation in generative AI. It’s up to startups to navigate these challenges without slowing the pace of technological progress. The decision highlights the need for clarity in copyright law, but it should serve as a guide rather than a barrier for ambitious AI entrepreneurs.
Andrei Pavel is a technology lawyer advising startups on IP, SaaS contracts, and compliance, helping them start, build, and scale across the EU. He has hands-on experience building digital tools and works with early-stage companies on product launches, funding, and regulatory strategy. Andrei also mentors founders in accelerator programs and deals with legal innovation projects at Ventures’n’Law.
