Hack reveals Suno trained AI music on millions of songs

Hacked Suno materials showed that the AI music generator had been trained on millions of songs and lyrics scraped from YouTube Music, Deezer and Genius, with additional data sources that included Pond5, Jamendo, Freesound, the International Music Score Library Project and podcast RSS feeds. The leak offered a rare look inside a training pipeline Suno had kept largely opaque, even as the company pushed its music tools into the commercial market.

The exposed files also suggested Suno pulled from decades of music and podcast material across the open internet, raising the same question at the center of the company’s legal fight with the recording industry: who gets to decide when copyrighted work can be copied for machine learning, and who gets paid when it is. The dispute is not simply about whether Suno scraped content. It is about whether AI firms can build profitable products on unlicensed media at massive scale while leaving artists, labels and other rightsholders to discover the use after the fact.

That legal battle is already well advanced. On June 24, 2024, the Recording Industry Association of America announced federal lawsuits in the United States District Court for the District of Massachusetts on behalf of Sony Music Entertainment, Universal Music Group and Warner Records, accusing Suno of mass infringement for copying copyrighted sound recordings to train its models without permission. The original complaint cited 560 copyrighted works. Later filings sought to add 61,026 recordings after the labels used audio fingerprinting to identify more material in Suno’s training data.

Suno’s co-founder and chief executive, Mikey Shulman, has defended the company by saying it trains on music found on the open internet and that its technology is transformative. He has said, “Suno’s mission is to make it possible for everyone to make music.” The company has also argued that its system is designed to generate new outputs rather than memorize and regurgitate existing songs.

The leak lands in the middle of a wider fight over AI accountability, where disclosure lags far behind commercial deployment and the music business is still trying to establish what consent, compensation and licensing should look like in an era of industrial-scale scraping.