Leaked internal data has revealed new details about how AI music platform Suno allegedly collected recordings to train its song-generation models.
According to a report published by 404 Media, the information came from a hacker who claimed to have breached Suno’s systems.
The leaked files reportedly date back to 2023 and 2024.
The documents suggest Suno gathered audio from multiple sources.
They reportedly include YouTube Music, Deezer, Genius, and stock audio libraries such as Pond5, Jamendo, and Freesound.
Alleged Training Methods Add to Legal Challenges
The latest report adds to the legal scrutiny surrounding Suno’s AI training practices.
The company is already facing a lawsuit from Hagens Berman, the law firm behind the largest litigation settlement in history.
Suno also faces a copyright lawsuit filed in 2024 by the Recording Industry Association of America (RIAA).
Sony Music, Universal Music Group, and Warner Music Group joined the legal action.
The lawsuit alleges that Suno trained its AI models using millions of copyrighted recordings without authorization.
Suno has not denied using copyrighted material for training. However, the company argues that its practices fall under the fair use doctrine.
According to 404 Media, the leaked files appear to support claims made in the RIAA lawsuit.
The report says internal metadata references several datasets containing more than two million audio clips from YouTube Music alone.
The leaked data also points to additional material collected from Genius, the IMSLP sheet music archive, and other online sources.
Together, those datasets reportedly contain decades’ worth of music and podcast recordings used to train Suno’s AI models.
Read more: Cheerful Music Explores AI and Human Creativity at AI Summit
Suno Allegedly Targeted Isolated Vocal Recordings
The leaked documents also suggest Suno specifically searched for isolated vocal recordings, including acapella versions of songs uploaded to YouTube.
According to the report, the company also relied on Bright Data’s web-scraping infrastructure and proxy services to download the material.
The leak comes as artists continue raising concerns about AI training datasets.
Those discussions intensified after The Atlantic released its AI Watchdog database.
The tool allows musicians to check whether their work appeared among more than 21 million files reportedly circulated to AI developers without permission.
The database does not directly mention Suno. Still, the company has drawn significant attention because of its ongoing copyright lawsuits.
Suno responded by saying it experienced what it described as a limited security incident in November.
The company added that it contained the incident quickly and later concluded that the exposed information mainly consisted of outdated source code that is no longer in use.