A new investigation and search tool from The Atlantic reveals which artists have had their catalogs harvested for AI training datasets.
The Atlantic’s 'AI Watchdog' project has launched a searchable database mapping millions of songs used to train generative models from Google and Stability AI. The tool allows artists to see if their tracks were ingested into datasets often licensed under Creative Commons, which technically bar commercial exploitation.
The findings are stark. High-profile acts like Skrillex, Peggy Gou, and Bicep have had dozens of tracks scraped, while underground producers are discovering their niche releases are also in the mix. SZA recently reported that over 200 of her songs, including unreleased material, surfaced in the data.
The backlash from the community is intensifying. Kenneth Blume (fka Kenny Beats) publicly called out AI generators for profiting off the labor of struggling musicians. While the database isn't exhaustive, it provides rare transparency into how tech companies are mashing up human creativity to power their algorithms.
For artists, this is a necessary audit. Check the AI Watchdog to see where your audio stands in the current training cycle.