ChatGPT’s Use of Song Lyrics Violates Copyright, Munich Court Finds

A Munich court ruled that ChatGPT, developed by OpenAI, unlawfully reproduced copyrighted lyrics in its AI models, which include GPT-4 and GPT-4o. The decision, which may set a significant precedent in Europe, orders OpenAI to cease reproducing these lyrics, disclose related training information, and compensate the rights holders. The court recognized that the models memorized specific lyrics from nine well-known songs, thereby satisfying the legal definition of copyright infringement under the EU InfoSoc Directive and German copyright law. OpenAI argued that its models do not store training data directly and generate outputs based on user prompts. However, the court rejected this argument, clarifying that full reproductions embedded in the model's structure are beyond the scope of data-mining exemptions. This ruling marks the first instance of a European court finding that a large language model has violated copyright through memorization, raising significant questions about how AI companies will source and license creative material in the future.

Source 🔗