The Munich District Court ruled on Tuesday that U.S. artificial intelligence company OpenAI's chatbot ChatGPT violated German copyright law by reproducing lyrics from songs by best-selling musician Herbert Gronemeier and others. The high-profile case focuses on whether OpenAI’s use of lyrics to train language models is legal.
The court found that OpenAI’s artificial intelligence system used the protected content of nine German songs as training data, including Groenemeier’s hits Maenner and Bochum.
The case was brought by the German Music Copyright Association (GEMA). Members of the association include composers, lyricists and music publishers, and the case has become another landmark event for global artists to fight back against artificial intelligence data scraping.
Presiding judge Elke Schwager ordered OpenAI to pay compensation for its use of copyrighted material, but did not disclose the specific amount.
Kai Welp, legal adviser to the German Music Copyright Association, said that the association hopes to start consultations with OpenAI in the future to clarify how copyright owners should receive reasonable remuneration.
OpenAI argued that its language model did not store or copy specific training data and could only reflect what was learned based on the overall training data set.
The company also contends that the chatbot's output is only generated after the user enters the prompt word, so the relevant legal liability should be borne by the corresponding user, not the defendant (OpenAI).
However, the court pointed out in its ruling that the language model’s memory behavior of lyrics and the reproduction of lyrics in the chatbot’s output content both constitute an infringement of the right to use copyright.