A New Public Tool Reveals Which Artists’ Songs are Being Used for AI Training
The tool draws from four primary databases that have millions of songs between them
The Atlantic has launched a new AI Watchdog tool to allow artists to see which of their songs are being used for AI training. All artists need to do is head to this page and type in their name, and a list of the songs will appear.
More information:
The tool draws from four datasets that share millions of songs among the AI development community.
One set contains 12 million songs. Another contains nine million, and the other two have over 100,000.
Three of these datasets scraped music from Spotify and YouTube.
The tool works for all forms of creativity, not just music.
Users can search for authors, YouTube creators, screenwriters, and actors.
According to the page, not all tracks that come up in the search have been used.
If a song doesn’t come up, it still may have been used in a different dataset.
Some songs are listed multiple times in the datasets.
Background:
Researcher Alex Reisner compiled the datasets for the tool.
He found them in research papers and AI data-sharing sites.
Google and Stability have used these data sets.
The numbers:
New Industry Focus looked at how many songs were found in the datasets from the artists with the top five monthly listeners on Spotify as of June 26, 2026.
The number below associated with “songs” is the total number of songs among the datasets found when each artist is searched in the tool:
Bruno Mars - 133.6m monthly listeners - 174 songs.
Justin Bieber - 128.4m monthly listeners - 506 songs.
The Weeknd - 113.9m monthly listeners - 488 songs.
Rihanna - 110.3m monthly listeners - 518 songs.
Michael Jackson - 109.6m monthly listeners - 622 songs.
👋 Disclosures & Transparency Block
This article was written with information sourced from Resident Advisor and The Atlantic.
We covered it because of the ongoing conversation around AI in music.
