AI music datasets contain millions of copyrighted songs, sparking legal battles

An investigation conducted by Alex Reisner of The Atlantic uncovered that four datasets utilized for training AI music generators comprise millions of copyrighted tracks, featuring songs from Italian musicians such as Vasco Rossi, Fabrizio De André, and Marracash. These datasets encompass 12 million MP3 files along with links to tracks on platforms like Spotify and YouTube. Developers circumvent protections to acquire audio. In 2022, both Google and Stability AI accessed one of these datasets. AI services, including Suno and Udio, create music based on prompts, imitating original compositions. In November, Czech skaters performed to an AI-generated piece with lyrics from New Radicals' "You Get What You Give." In 2024, major labels filed a lawsuit against Suno and Udio for copyright violations, demanding $150,000 for each song. Suno's subscription is $8 per month, and in September, Spotify removed 75 million AI-generated songs.