AI Music Database Exposed
Your favorite artists are powering AI. 🎵 The Atlantic just launched a searchable database exposing 21 million tracks used for AI training. Here is how they did it.
What the video says
Over 21 million tracks of copyrighted music have been exposed in a tipping point for the industry, as The Atlantic launches a historic, first-of-its-kind searchable database of AI training data. Reporter Alex Reisner uncovered 4 massive datasets being used to train AI models, featuring artists ranging from Lady Gaga and Radiohead to Wu-Tang Clan.
The 2 largest collections contain 12 million and 9 million tracks respectively, while 2 smaller sets hold over 100,000 songs each. This never-before-seen transparency has forced a reckoning, with tech giants Google and Stability already confirming in research papers that they used these very datasets.
While the sets are technically public, Reisner revealed that 3 of them are actually just links to Spotify and YouTube. To build their models, developers use automated tools to scrape the audio, bypassing logins, ads, and creator monetization in direct violation of platform terms of service.
Now, the public can search this flagship database directly on The Atlantic's AI Watchdog site to see exactly which artists are powering the world's AI models.
