The Regional Court of Munich’s November 2025 decision in GEMA v OpenAI marks a defining moment for the global conversation around copyright and AI. The court held that OpenAI’s ChatGPT had violated German copyright law by reproducing song lyrics from GEMA’s catalogue without proper licensing. In my opinion this ruling resonates far beyond Germany, signalling that in the age of generative AI, creators’ rights remain firmly protected. [i]

Background

GEMA, which represents composers, lyricists, and music publishers in Germany, brought the case over nine of the country’s best-known songs. These included Herbert Grönemeyer’s 1984 hit Männer and Helene Fischer’s Atemlos Durch die Nacht, which was the unofficial anthem of the German side during the 2014 Football World Cup. [ii]

When prompted by users, ChatGPT generated full or nearly complete versions of these copyrighted lyrics, showing that OpenAI’s system had “memorised” the works during training. OpenAI’s lawyers argued that its model did not store copyrighted text but generated new material based on statistical patterns. It also invoked the EU’s “text and data mining” (TDM) exception[iii], which allows limited use of data for machine learning. The court rejected this defence, concluding that reproducing song lyrics went beyond permissible data mining and amounted to infringement. As a result, GEMA was entitled to the claims for injunctive relief, information and damages.[iv]

Significance

This ruling should reshape how AI companies interact with creative industries. It establishes a clear precedent that “memorisation” within an AI system, where models can regenerate copyrighted works, may constitute reproduction under copyright law.

This is one of the first European judgments to hold an AI developer liable not for traditional copying, but for what its model learned and could later output. The court’s reasoning will likely influence how AI developers manage training data and licensing obligations. Developers may now need to obtain explicit permissions for copyrighted content used to train generative models, particularly music, literature, and art. The ruling may also push AI developers to disclose what data their models use, promoting transparency but also adding compliance challenges.

Interestingly, this strict interpretation contrasts with the UK’s Getty Images v Stability AI case[v]. There, the High Court found that Stability AI’s model had not infringed copyright because it learned from data statistically rather than storing it.[vi] This divergence highlights a growing legal gap between the EU and UK courts. While the UK leans toward a permissive view of machine learning, the EU appears to prioritise creators’ rights – even when innovation is at stake.

Key Takeaways

• Memorisation equals reproduction: If an AI model can regenerate protected works, that may amount to infringement.
• Dual exposure: Liability can arise both during training (embedding copyrighted works) and generation (reproducing them).
• Licensing obligations: Developers should secure rights for creative data sets before training.
• Jurisdictional divergence: Europe’s stricter approach diverges from the UK’s more flexible stance, creating compliance complexities for global developers.

Conclusion

GEMA v OpenAI [vii] underscores a pivotal shift in how courts interpret AI’s relationship with copyright. By recognising that training an AI can infringe when memorisation leads to faithful reproduction, the Munich court has drawn new lines in the sand. For AI developers, innovation must now consider alignment with established copyright principles. Whitestone will certainly working with it’s creative clients in music and the arts to look further into protecting their intellectual property rights.

© Lawrence Power 2025

---

[i] GEMA v OpenAI (Munich District Court I, 11 November 2025) < https://media.licdn.com/dms/document/media/v2/D4E1FAQGoTQ9DKfW1uw/feedshare-document-pdf-analyzed/B4EZp0_lp_IwAc-/0/1762899432336?e=1764201600&v=beta&t=HborRwy3RYuqCH4P0PabjE-jXUYKK7My5_vCGX7pTmg> accessed on 13 November 2025.

[ii] The Guardian, ‘ChatGPT violated copyright law by ‘learning’ from song lyrics German court rules’ , 2025, <https://www.theguardian.com/technology/2025/nov/11/chatgpt-violated-copyright-laws-german-court-rules > accessed on 13 November 2025.

[iii] TDM, exception cite

[iv] GEMA v OpenAI (Munich District Court I, 11 November 2025) <  > accessed on 13 November 2025.

[v] Getty Images v Stability AI [2025] EWHC 2863 (Ch).

[vi] Getty Images v Stability AI [2025] EWHC 2863 (Ch) [554].

[vii] GEMA v OpenAI (Munich District Court I, 11 November 2025) <  > accessed on 13 November 2025.