The Atlantic Reveals Databases of Millions of Songs Used for AI Training

The Atlantic has released four searchable databases that reveal the extensive use of music recordings in training artificial intelligence models. One database contains 12 million tracks, another 9 million, and the remaining two each hold about 100,000 songs. Staff writer Alex Reisner highlights that the databases include copyrighted music from artists such as Taylor Swift and Bad Bunny.

The publication comes amid ongoing legal battles against AI music platforms Suno and Udio, which have invoked fair use as a defense against claims of unauthorized scraping of copyrighted content. A similar case in book publishing saw a judge reject copyright infringement claims, but piracy allegations proved more compelling. An initial settlement of $1.5 billion was reached, though the final outcome remains pending.

These databases could assist the music industry in pursuing similar lawsuits in the future. Many streaming services have implemented measures to prevent, identify, or label AI-generated content, but with varying success. They have also failed to stop scammers from using AI to create imitations of established bands and profit from their work.