Hacked Suno data points to training scrape from music sites

Hacked data from Suno indicates the AI music generator trained on large collections of songs and lyrics taken from online services, 404 Media reported. The report matters because Suno has faced lawsuits over whether it used copyrighted works to build its music models.

404 Media reported that the data came from a hacking incident and showed material scraped from platforms including YouTube Music, Deezer and Genius. The Verge reported that the material included millions of songs and lyrics.

The report gives an unusual look at Suno’s training data, The Verge reported. Suno has not publicly detailed what sits inside its training datasets or how the company obtained the material used to develop its systems, according to The Verge.

Suno’s data practices are already under legal scrutiny. The Verge reported that multiple lawsuits accuse the company of using copyrighted material to train its AI models.

One prominent case was filed by the Recording Industry Association of America, The Verge reported. The RIAA case is part of a wider fight over how copyright law applies when AI systems learn from existing music and then generate new audio.

The hacked-data report does not resolve the legal claims against Suno. It does, however, add specific platform names to the public record, with 404 Media identifying YouTube Music, Deezer and Genius as sources reflected in the data.

For music companies, artists and online platforms, the report sharpens a central dispute around generative AI: whether companies can train music models on large stores of existing songs and lyrics without permission. For Suno, it brings new attention to a question the company has avoided answering in detail, The Verge reported: what its models were trained on and where that material came from.

This story draws on original reporting from The Verge.