Germany Fines OpenAI for Illegally Using Song Lyrics to Train ChatGPT (techoreon.com)

A Munich court has ruled that OpenAI violated German copyright law by including song lyrics in the training data for its language models and ordered the company to pay compensation to writers. The case, brought by German rights collective GEMA, concerned nine German-language songs whose lyrics appeared verbatim in ChatGPT outputs. Judges found that both the internal “memorisation” of those texts in model parameters and their later reproduction in chatbot responses constituted copyright infringement. The court rejected OpenAI’s defenses—that models don’t store specific works and that users, not the company, are responsible for outputs—and held the company liable; OpenAI says it disagrees and is considering next steps. The decision is significant because it’s billed as a major European precedent on using copyrighted material to train generative AI, and could influence similar lawsuits in the U.S. and licensing negotiations worldwide. Technically, the ruling treats model parameter “memorisation” and verbatim regeneration as actionable copying, increasing legal risk for training on unlicensed copyrighted content. Practically this raises pressure on AI developers to adopt licensed datasets, stronger data provenance, filtering or memorisation-limiting training methods, and output controls (e.g., detection/watermarking or refusal to reproduce exact copyrighted passages). The verdict signals that courts may hold platform builders accountable for both the training pipeline and model outputs, reshaping data governance and commercial licensing strategies for foundation models.