The Atlantic exposes the music datasets powering AI—and makes them searchable

What happened

Atlantic reporter Alex Reisner identified four datasets of music being used to train AI models and made them fully searchable for the public. Two of the sets are enormous—one containing 12 million tracks and another 9 million. The remaining two are smaller but still substantial, each holding over 100,000 songs.

According to Reisner, the datasets have been downloaded thousands of times. While it's impossible to know exactly who has used them, both Google and Stability AI have confirmed in research papers that they trained on some of this data.

Three of the four datasets are distributed not as audio files but as lists of links to songs hosted on YouTube or Spotify. AI developers then use automated tools to download the actual audio—tools that can bypass logins, advertisements, and other mechanisms designed to compensate creators. As Reisner notes, such tools violate the terms of service of those platforms.

Some sources, like the Free Music Archive dataset, are free to stream for personal use but require licensing for commercial applications—raising questions about whether AI training constitutes commercial use.

Why it matters

This is the first time the public can search and verify whether specific songs or artists appear in the training data behind AI music models. That transparency matters for several reasons:

• Artist rights: Musicians and rights holders can now check if their work was used without consent or compensation.
• Legal exposure: The datasets' reliance on scraping YouTube and Spotify links—potentially violating platform terms of service—adds fuel to ongoing litigation around AI training data.
• Industry precedent: With Google and Stability AI already confirming use, the database provides concrete evidence for lawmakers, regulators, and courts examining how AI companies source creative works.
• Public accountability: Making the data searchable shifts the burden of transparency from AI companies (who have largely stayed silent) to the public, empowering journalists, researchers, and artists to investigate independently.

The scale is striking: over 21 million tracks across the four datasets, downloaded thousands of times by parties that may include major tech companies.

Public reaction

No strong public signal was available from Reddit or other discussion forums at the time of this report. The story is still developing, and community discussion may emerge as artists and rights holders begin searching the database for their own work.

What to watch

• Whether artists or labels use the searchable database to file new lawsuits or join existing ones.
• How Google and Stability AI respond to the confirmed use of these datasets in their research.
• Whether other AI companies disclose their training data in response to growing public pressure.
• Potential regulatory action, particularly in jurisdictions where scraping streaming platforms for commercial AI training may violate data or copyright laws.
• Whether The Atlantic's database prompts similar investigative efforts into training data for other modalities (text, image, video).

Sources

• The Verge — The Atlantic created a searchable database of the music used to train AI
• Wilson's Media — The Atlantic created a searchable database of the music used to train AI
• GNNHD — The Atlantic created a searchable database of the music used to train AI
• Amkio — The Atlantic created a searchable database of the music used to train AI