Suno Secretly Scraped Millions of Songs From YouTube, Genius, and Deezer to Train Its AI

If you've ever uploaded a song to YouTube, posted lyrics to Genius, or streamed anything on Deezer, there's a decent chance your music ended up in Suno's training data without your knowledge or consent. A new report from The Verge reveals that the AI music generation startup scraped millions of songs from all three platforms to build the models that power its product.

This is a significant story, and not just because it involves a well-known startup doing something legally murky. It matters because Suno has been one of the most visible faces of the AI music boom, generating enormous enthusiasm (and investment) on the promise that it can create studio-quality tracks from a text prompt. That promise looks rather different when you understand the foundation it was built on.

What Was Actually Taken

The scope here is hard to overstate. YouTube alone hosts hundreds of millions of tracks, from major label releases to bedroom producers who uploaded their first demo at 17 and have long since forgotten about it. Genius is the dominant platform for song lyrics, with a catalogue that spans virtually every genre and era of popular music. Deezer is a streaming service with a library of over 90 million tracks.

Suno's scraping operation pulled from all three. The exact number of works affected has not been confirmed by the company, but descriptions of the scale suggest it runs into the tens of millions of individual recordings and associated text data.

This is not the first time an AI company has faced these accusations. Stability AI, Midjourney, and OpenAI have all had legal challenges filed over their training data practices. But the Suno situation is notable because the sourcing appears to have been both broad and deliberately opaque. There was no public disclosure, no opt-out mechanism offered to creators, and no licensing deals struck with the platforms involved.

The Legal and Ethical Tangle

From a legal standpoint, Suno's position is precarious. The major record labels, including Universal Music Group, Sony Music, and Warner Music Group, filed a copyright lawsuit against Suno in mid-2024. That case is still working through the courts, and revelations about the specific sources of training data are unlikely to help the company's defence.

The argument that AI companies often make, that scraping publicly available data for training constitutes fair use under US copyright law, has not yet been definitively tested in court. But the more specific the sourcing becomes, the harder that argument is to sustain. Scraping a general snapshot of the internet is one thing. Deliberately targeting a music streaming service's catalogue is quite another.

For independent artists, the frustration runs deeper than legal theory. Many of them chose platforms like YouTube specifically because it offered some visibility and control over their work. Finding out that a commercial AI product was trained on their output, and is now generating music that competes directly with them commercially, feels like a profound betrayal.

What Suno Has Said

As of publication, Suno has not issued a detailed public response to the specific sourcing allegations. The company has previously argued that its training practices are lawful and that its outputs are sufficiently transformative to qualify as fair use. That position is going to be tested hard in the months ahead.

YouTube's terms of service explicitly prohibit scraping without permission. Deezer and Genius have similar restrictions. Whether those terms of service violations translate into additional legal liability, on top of the copyright claims, is something courts will need to work out.

The Bigger Picture

The music industry has historically been faster than other creative sectors to fight back against what it sees as unauthorised use of its catalogues. The Napster and streaming-era battles gave labels the experience and legal infrastructure to move quickly. AI is the newest front, and Suno is now very much in the centre of it.

For listeners and fans of AI-generated music, none of this necessarily changes the experience of using the product. But it should prompt a harder question about what it means to enjoy a tool that may have been built on work taken without permission. The technology is impressive. The ethics of how it got there are considerably less so.