The Case That’s Shaking the AI Industry
On November 11, 2025, the Munich Regional Court issued a decision* that could have far-reaching consequences for the development of artificial intelligence. GEMA, Germany’s largest music rights management organization, had sued OpenAI – and won. The allegation: ChatGPT had used copyrighted song lyrics without a license and reproduced them almost verbatim.
*LG München I, Endurteil v. 11.11.2025 – 42 O 14139/24– can be found here: gesetze-bayern.de
The Core Question: What is Text & Data Mining?
In its defense, OpenAI had invoked so-called Text & Data Mining (TDM). This exception anchored in copyright law generally allows the use of protected works for automated analysis of large data sets. The idea behind it: science and research should be able to recognize patterns and correlations in texts without having to acquire a license for each individual text.
However, the Munich court made it clear: What OpenAI was doing goes beyond pure TDM. The crucial difference lies in the type of use. TDM typically analyzes data and derives new insights from it. ChatGPT, however, had apparently processed the song lyrics in such a way that they could later be output almost word-for-word.
The Problem of “Memorization”
A central concept in this legal dispute is so-called “memorization.” This refers to the process by which AI systems go beyond simply learning patterns from training data. Instead, they effectively store this content and can later reproduce it.
In large language models like ChatGPT, memorization can occur under specific conditions. This happens when certain texts are processed so frequently during training or are so distinctive that the model essentially “memorizes” them. The system then develops the ability to reproduce these texts almost identically. This process is similar to a human who has memorized a poem.
It was precisely this memorization that became OpenAI’s undoing. The court’s argumentation addressed a crucial distinction in AI usage. When an AI internalizes copyrighted works so strongly that it can reproduce them verbatim, it crosses a legal boundary. This is no longer permissible analysis, but constitutes unauthorized reproduction and public communication.
The Legal Classification
The Munich Regional Court established a clear boundary in AI copyright law. When AI training results in content being permanently retained or later reproduced word-for-word, it exceeds the scope of the TDM exception. This exception is intended only for pure analysis, not for the reproduction of protected works.
Therefore, if an AI application uses and reproduces song lyrics without a license, developers violate German copyright law. This directly interferes with the economic exploitation interests of the creators. Users could then access texts via AI instead of purchasing them from licensed providers.
Reactions and Outlook
Dr. Tobias Holzmüller, CEO of GEMA, was combative after the verdict: The internet is not a self-service store and human creative achievements are not free templates. They had created a precedent that makes clear – operators of AI tools must also comply with copyright law. Therefore GEMA sees the verdict as a successful defense of musicians’ livelihoods (Zitat: gema.de)..
The verdict is not yet legally binding. OpenAI could appeal, and it remains to be seen whether higher courts will share the Regional Court’s assessment.
What Does This Mean for the Future?
This decision could have profound implications for the development and deployment of AI systems. Companies may need to rethink their training methods and ensure that their models cannot memorize and reproduce protected content. This could lead to technical adjustments, such as filters that prevent copyrighted texts from being output.
At the same time, the verdict raises fundamental questions: How can AI developers ensure that their systems do not violate copyrights? What technical measures are necessary and practicable? And how can innovation in AI development be reconciled with the protection of creative works?
In any case, the Munich verdict makes one thing clear: The message “Copyright remains in force – even in the AI age” has reached the courts. Song lyrics and other creative works are not free training resources for artificial intelligence. Those who want to use them must pay for it – or find technical ways that exclude memorization and reproduction.
You want to read more about AI? Check out our articles on “What is an AI System?” and “CRA vs. AI Act“.
Do you have questions about using AI systems? Secure your free initial consultation now!
