Over 21 Million Copyrighted Songs Found in AI Training Datasets
A recent investigation by The Atlantic has unveiled that over 21 million copyrighted songs are currently circulating among AI developers, serving as the backbone for training generative music systems. The report, led by writer Alex Reisner, highlights that developers have gained access to four major music datasets, which include works from both independent creators and global superstars such as Billie Eilish, Taylor Swift, Nirvana, and Bad Bunny.
The scale of these datasets is staggering. While two of the collections contain over 100,000 recordings, the remaining two are significantly more expansive, housing between 9 million and 12 million tracks each. Despite the widespread use of these files—with thousands of downloads recorded—the lack of transparency in AI training practices makes it difficult for artists to track exactly which companies are utilizing their intellectual property. However, the report specifically identifies Google and Stability AI as entities that have leveraged the Free Music Archive dataset for training purposes.
The Launch of the AI Watchdog Tool
To provide transparency to the creative community, The Atlantic has introduced an AI Watchdog tool. This resource allows musicians to search by artist name to determine if their work is included in the four identified datasets. Early findings from the tool have revealed significant presence of specific artists, including Björk (411 songs), Moby (213), Fatboy Slim (175), The Chemical Brothers (153), Daft Punk (151), and Charlotte de Witte (89), among many others.
The reaction from the music industry has been swift and condemnatory. SZA took to social media to express her frustration, noting that 238 of her songs were used for training, including unreleased material. She described the practice as “degenerate,” echoing a growing sentiment among artists who feel their life’s work is being exploited without consent or compensation.
Industry Backlash and Legal Battles
Producer Kenny Beats directed his criticism toward AI music company Suno, questioning the ethics of a business model built on the unauthorized use of creative labor. Similarly, DJ Sabrina the Teenage DJ highlighted the irony of her music being used to train AI, noting that her own unique sound had previously been unfairly criticized as “AI slop” by listeners unaware of the underlying data usage.
This revelation arrives amidst a broader climate of legal tension. Major AI music companies, including Suno and Udio, are currently facing multiple lawsuits from record labels, unions, and individual artists. While Universal Music recently settled its own legal dispute with Udio—resulting in a partnership to develop a new platform—the industry remains deeply divided. As AI companies continue to defend their practices, the debate over the ethics of generative music and the protection of human artistry shows no signs of slowing down.
