Leaked Source Code Reveals Suno AI Trained on Scraped Music from YouTube, Deezer, and Genius
A recent security breach has exposed details about how Suno, an AI music generation company, obtained training data for its model. Hacked source code shows that Suno scraped decades worth of music, podcasts, and associated lyrics from multiple platforms, including YouTube, Deezer, and Genius.
The leak provides a rare glimpse into the data practices of AI companies that often keep their training methodology confidential. According to the exposed code, Suno appears to have systematically collected audio and text content from these services to build a comprehensive training dataset for its music-generating AI.
This revelation arrives amid ongoing legal challenges faced by AI music companies. Major record labels have filed lawsuits against Suno and competitors like Udio, accusing them of using copyrighted material without permission to train their models. The companies have defended their practices, though this latest disclosure may complicate those arguments.
The incident underscores the broader debate surrounding AI training data. Critics argue that scraping content from the open web without explicit consent constitutes a form of copyright infringement, while AI developers often contend their use falls under fair use principles.
It remains to be seen how this exposure will impact Suno's ongoing legal battles or influence future regulations around AI training practices.