OpenAI loses song lyrics copyright case in German court (www.dw.com)

A Munich court ruled that OpenAI’s ChatGPT infringed German authors’ rights by using song lyrics in responses without paying license fees, ordering damages though not specifying an amount. Judge Elke Schwager found that GEMA — the German association representing songwriters — proved ChatGPT had been trained on nine specific songs (including Herbert Grönemeyer’s "Männer" and Rolf Zuckowski’s "In der Weihnachtsbäckerei") and that the model’s reproduction of lengthy lyrics could not plausibly be coincidence. OpenAI conceded the lyrics were part of its training data but argued LLMs do not store verbatim texts, only learned parameters, and that outputs are prompted by users; the court rejected these defenses. OpenAI says it will appeal and is negotiating with rights organizations. The ruling is significant because it invokes Germany’s authors’ rights (Urheberrecht), a more artist-centric regime than Anglo-American copyright, and could influence EU-wide practice since relevant rules are largely harmonized. Technically, the decision challenges the industry’s claim that parameterized models absolve providers from liability for verbatim reproductions and may push companies toward explicit licensing, stricter data curation, or new filtering for copyrighted content. Media and journalists’ unions hail the verdict as strengthening creators’ claims against AI training practices, potentially prompting broader licensing talks and litigation across Europe.