Which companies use these music databases for AI training?

Google and Stability AI have confirmed using such datasets in their research. However, the full list of companies remains unknown, as the databases have been downloaded thousands of times.

Why does the use of this data spark controversy?

Many tracks are distributed via YouTube and Spotify without adhering to licensing agreements. Automated tools bypass platform protections, depriving creators of revenue and violating service terms.

Can these datasets be legally used for commercial projects?

No, most tracks require licensing for commercial use. For example, the Free Music Archive allows free listening but prohibits commercial use without rights holders' consent.

← All news

Artificial intelligence

The Atlantic Publishes Music Database for AI Training

June 21, 2026

Photo: The Verge

Quick answer

Journalists from The Atlantic released a searchable database of 21 million music tracks used for AI training.

Journalist Alex Reisner from The Atlantic published an investigation revealing four major datasets used to train artificial intelligence models. The largest collections contain 12 and 9 million music tracks, while the other two datasets include over 100,000 compositions each. According to Reisner, these databases have been downloaded thousands of times, with companies like Google and Stability AI confirming their use.

Despite the availability of these datasets online, their use involves technical and legal challenges. Three of the four datasets consist of links to tracks on YouTube and Spotify rather than ready-made audio files. Developers use automated tools to download music, bypassing platform protections such as authentication and ad blocks. This violates service terms and deprives content creators of potential revenue.

Legal aspects also raise concerns. For instance, the Free Music Archive permits free listening but requires licensing for commercial use. Using such data without rights holders' consent could lead to legal disputes, especially if AI models generate music resembling copyrighted works.

Common questions

Which companies use these music databases for AI training?: Google and Stability AI have confirmed using such datasets in their research. However, the full list of companies remains unknown, as the databases have been downloaded thousands of times.
Why does the use of this data spark controversy?: Many tracks are distributed via YouTube and Spotify without adhering to licensing agreements. Automated tools bypass platform protections, depriving creators of revenue and violating service terms.
Can these datasets be legally used for commercial projects?: No, most tracks require licensing for commercial use. For example, the Free Music Archive allows free listening but prohibits commercial use without rights holders' consent.

Dzen feed: /feed/dzen.xml · RSS: /feed.xml