The Munich Regional Court ruled on Tuesday that OpenAI's chatbot ChatGPT violated German copyright law by training its artificial intelligence using lyrics copied from hit songs by musician Herbert Gr?nemeyer and others.

Presiding Judge Elke Schwager ordered OpenAI to pay damages for its use of copyrighted material, though the specific amount was not disclosed.

Kai Welp, legal counsel for GEMA, stated the organization now seeks discussions with OpenAI on how copyright holders should be compensated.

The case was brought by German music rights association GEMA, whose members include composers, lyricists, and publishers.

OpenAI had argued its language model neither stores nor replicates specific training data, but rather reflects what it learned from the entire training dataset. OpenAI contends that since outputs are solely the result of user input (i.e., prompts), liability rests with the respective users rather than the defendant.

However, according to the court ruling, the court determined that OpenAI's ChatGPT infringed copyright by using and memorizing copyrighted song lyrics during training and subsequently generating these lyrics in response to simple user prompts. Consequently, the court ruled against OpenAI, ordering it to cease infringement, provide information on infringing activities, and compensate for damages.

The core legal basis for this ruling is that an AI model's “memorization” of training data and its ability to reliably reproduce it constitutes ‘reproduction’ under copyright law. Furthermore, the model's output of lyrics based on simple prompts constitutes “distribution” to the public.

Furthermore, the court explicitly rejected OpenAI's defense invoking the “text and data mining” exception, ruling that this exception applies only to temporary reproductions for analytical purposes and does not cover the new exploitative act of permanently storing works within the model for generating outputs.

Earlier this year, major Bollywood music companies petitioned a New Delhi court to join a copyright lawsuit against OpenAI, alleging unauthorized use of recorded materials to train AI models.

Scan QR code for original text:

Scan QR code for translation: