A German music rights body is suing OpenAI for copyright infringement, claiming the US tech giant’s AI systems reproduced several artists’ lyrics without permission, a court heard Monday.
It is the latest court case that OpenAI has faced, with media groups and authors among those claiming that the company’s ChatGPT chatbot has been trained on their work without permission.
The music rights organisation GEMA claimed that OpenAI had “systematically” used its repertoire to train its artificial intelligence models, according to a statement from a regional Munich court which is hearing the case.
GEMA — which has more than 100,000 composers, songwriters and publishers as members — is representing lyricists of nine German songs in the case, and is seeking unspecified damages.
The court said it was “undisputed” that OpenAI’s large language model was trained with the lyrics from the songs.
In response to simple prompts, its chatbot reproduced “large parts of the song lyrics faithfully”, it said.
According to GEMA, this proves the lyrics are memorised in OpenAI’s large language model, which it uses to train its chatbot.
GEMA believes this means copyright has been infringed as permission was not first sought to use the lyrics, the court said.
OpenAI disagrees however, according to the court.
The company says its large language models do not store or copy specific data but rather reflect in their settings what they have learnt, according to the court.
With regard to the AI chatbot, it is users who are the producers of its output and are responsible for it, OpenAI claims.
The Munich court will rule on the case on November 11.
