Urteil GEMA gegen Open AI:
> Sowohl durch die Memorisierung in den Sprachmodellen als auch durch die Wiedergabe der Liedtexte in den Outputs des Chatbot lägen Eingriffe in die urheberrechtlichen Verwertungsrechte vor
https://www.justiz.bayern.de/gerichte-und-behoerden/landgericht/muenchen-1/presse/2025/11.php
Bavaria court rules against Open AI as it can regurgitate the lyrics of songs like “Atemlos”
> Both the memorization in the language models and the reproduction of the song lyrics in the chatbot's outputs constitute infringements of copyright exploitation rights.
(Google Translated) https://www-justiz-bayern-de.translate.goog/gerichte-und-behoerden/landgericht/muenchen-1/presse/2025/11.php?_x_tr_sl=auto&_x_tr_tl=en&_x_tr_hl=en-US&_x_tr_pto=wapp
The Guardian also reports on the “atemlos” ruling. Surprisingly the Guardian doesn’t mention “memorization”, unlike the judges, who explicitly talk about memorization leading to copyright violation.
In our own work, we researched memorization in language models for code and ways to let them regurgitate training data:
> From the training data that was identified to be potentially extractable we were able to extract 47% from a CodeGen-Mono-16B code completion model.
> We also observe that models memorise more, as their parameter count grows, and that their pre-training data are also vulnerable to attack
En ook op de NOS: “ChatGPT schendt auteursrecht van 'Atemlos' en andere Duitse songteksten”
@avandeursen „Because its output is generated by users of the chatbot via their prompts, OpenAI said, they were the ones who should be held legally liable for it – an argument rejected by the court.“
Really? This was the line of arguing? Really? So “you’re holding it wrong”? All the AI couldn’t come up with a better defense?
Arie van Deursen
Michael Dorner