Atlantic Releases Database of Music Datasets Used to Train AI Models
The Atlantic created a searchable database containing four datasets of music used to train artificial intelligence models, revealing the scale and composition of training data used by AI music generators.
The Atlantic created a searchable database of four datasets of music used to train AI models. Two of the datasets contain 12 million and 9 million tracks respectively. According to The Verge, the other two datasets each contain over 100,000 songs. The datasets have been downloaded thousands of times. According to Musically.com, the datasets include copyrighted music that is not supposed to be freely available. According to Aimusicpreneur.com, the datasets include hit songs from Taylor Swift, Bad Bunny, and other major artists, and map copyrighted music behind AI music generators including Suno, Udio, and Google. Google and Stability have confirmed they have used the datasets in research papers, according to The Verge. Universal Music Group and Sony Music sought to add more than 61,000 recordings to their copyright lawsuit against Suno, according to Musically.com. Suno has been opposing the addition of more than 61,000 recordings to the copyright lawsuit, according to Musically.com. AI companies have defended their use of copyrighted music by citing fair use.
This account was written only from facts that survived Augur's
corroboration pass — 2 corroborated across opposed news blocs,
0 contested (attributed to both sides), 8
single-source (attributed). Nothing was added; no significance was inferred.
Model Qwen3-Next-80B-A3B-Instruct.
See the evidence & the verbatim quotes →