AI and Copyright: Why a ruling from Munich changes everything

What was going on?

GEMA, the German copyright organization, claimed that OpenAI had used song lyrics from the GEMA repertoire (including the well-known "Atemlos" by Helene Fisher) without a license while training its models. But that's not all; users could request complete or almost complete song lyrics with simple prompts. According to GEMA, this was a double infringement: both storing lyrics in the model (known as "memorization") and reproducing them in the output were not permitted, in its view.

OpenAI defended itself by arguing that its models do not "store" texts, but merely learn statistical patterns. Furthermore, it claimed that responsibility for the generated output lies with the user, who determines the outcome by means of the prompt. OpenAI also invoked the European exception for text and data mining.

What does the court say?

The court did not agree with OpenAI's arguments. It found that the song lyrics were reproducible in the models and that this constituted a form of impermissible reproduction within the meaning of copyright law.

The output—the display of texts on request—also constitutes an infringement. Important: according to the German court, the exception for text and data mining does not provide carte blanche in this case. Training AI with protected works without permission does not fall under this exception.

Why is this new and remarkable?

1. First European ruling on AI training and copyrightThis is the first time that a European court has explicitly ruled that training generative AI with protected works is a copyright-related act that requires the permission of the copyright holder, and therefore a license.
2. Memorization recognized as legally relevantThe court introduces the concept of "memorization" as a form of storage equivalent to reproduction. This provides a legal interpretation and framework for a technical phenomenon.
3. Limits to AI developmentThe ruling makes it clear that innovation does not take precedence over the law. AI companies will have to develop licensing models similar to how streaming services once decided to make agreements with rights holders.

What does this mean for the future?

This ruling is not only a victory for GEMA, but also a signal to the entire AI industry: the rules of the game are changing. Expect more lawsuits, stricter compliance requirements, and possibly new European regulations. For creative makers, this is a step toward protecting their works in the AI era. For tech companies, it means it's time to rethink their business model.

The ruling does not stand entirely on its own. Interestingly, the American AI company Anthropic recently decided to pay a group of authors $1.5 billion to settle a lawsuit they had filed because Anthropic had used their works without permission to train its chatbot. So Anthropic seems to be anticipating changes to the rules of the game.

Has the final word been said on this matter? Probably not. OpenAI has considerable financial resources at its disposal to appeal cases like this. This has now also been announced.

This ruling is a wake-up call. AI is no longer a legal gray area; the rules of the game seem to be slowly becoming clearer.