Suno trained its AI on two million tracks scraped from YouTube

Suno, the AI music generator that composes a track from a simple description, has been hacked, and the breach exposed the complete inventory of what was used to train it. It reveals millions of tracks scraped from all over the place, with YouTube and Deezer leading the pack.

The hacker apparently carried out a so-called supply chain attack in November - compromising a supplier first in order to recover an employee's credentials and work their way up to the source code. The whole thing was revealed by the American outlet 404 Media.

And the list is staggering. On YouTube Music alone, the inventory counts more than two million clips, totaling around 114,000 hours of audio. Add to that nearly 12,000 hours scraped from Deezer, 18,000 from the lyrics site Genius, 62,000 from the sound library Pond5, and - most notably - nearly one million hours pulled from some 420,000 podcasts collected directly via their RSS feeds.

The clever part is the method. To get clean vocal recordings, Suno specifically targeted a cappella versions - voice only, no instruments - that users post on YouTube.

And to fly under the radar of YouTube's anti-scraping protections, the company routed all of it through an intermediary - the firm Bright Data, which rents out addresses that mask where the requests are actually coming from.

The wildest part is that Suno had never really hidden what it was doing. The company had already acknowledged training on "essentially all music files of reasonable quality accessible on the internet" - tens of millions of recordings. Its defense, like every other generative AI, is fair use, the American copyright exemption.

Except this leak changes everything. Until now, the major labels - Universal and Sony, who are already suing Suno - could only put forward suspicions. Now it's Suno's own internal inventory naming the platforms one by one, hour by hour.

And deliberately circumventing YouTube's protections is no minor detail - it's potentially illegal under the DMCA, the major US law governing online copyright.

It's hard to plead fair use when your own documents lay out the plundering hour by hour. This leak could end up costing Suno dearly.

Then join my community on Patreon for exclusive articles, advanced tutorials and lots of other surprises I keep for my supporters. It's thanks to you that I can keep sharing my passion after 20 years!