One of the largest publicly known AI music training datasets has been taken offline, while a wave of class-action lawsuits against generative music companies moves forward in U.S. courts and a new academic paper frames the data-scraping debate in neo-colonial terms.
Dataset Takedown and Researcher Fallout
An investigation published last month identified four publicly circulating datasets containing approximately 21.2 million copyrighted recordings. A companion search tool allowed artists to check whether their songs were included, prompting dismay among musicians who found their work in the collections.
Three of the four datasets contain no audio files, only links to YouTube and Spotify paired with metadata. Developers typically feed these into automated ripping tools that bypass login screens, advertisements, and royalty payment mechanisms. The only confirmed users are Google and Stability AI, which cited the Free Music Archive, the smallest of the four collections, in published research papers. It remains unknown who downloaded what from the two datasets that account for the vast majority of the 21 million tracks, as no legislation currently mandates disclosure.
In the wake of the investigation, Sleeping-DISCO-9M, a dataset of just over 9.7 million tracks scraped from commercial music, was removed by its publisher, the Sleeping AI Research Collective. The takedown of the original download link does not guarantee the data has stopped circulating, and the dataset remains searchable through a publicly available tool.
A paper accompanying the dataset, titled “SLEEPING-DISCO 9M: A Large-Scale Pre-Training Dataset for Generative Music Modeling,” was also quietly withdrawn. Its abstract noted that “there are no open-source high-quality datasets representing popular and well-known songs.” The authors had intended the dataset for pre-training generative models. While the removal is not an admission of liability, plaintiffs in copyright cases could argue it shows awareness of potential legal risks. Without such datasets, researchers building open-source music models are pushed back toward royalty-free libraries, Creative Commons music, licensed catalogs, or synthetic datasets, all of which carry limitations in scale or musical diversity.
Sleeping AI issued a statement describing the “absolute erasure” as a deliberate decision “in response to emerging public controversies, localized targeting behaviors, and escalating incidents of doxing that compromised the personal security, privacy, and individual liberties of independent authors.” The statement denied that the dataset hosted, indexed, or distributed copyrighted multimedia content, adding that “nor were any generative AI models ever trained on this metadata.”
Class Actions and Licensing Fractures
Separate from the dataset controversy, a series of class-action lawsuits filed by independent musicians is progressing against AI music companies Suno, Udio, Mureka, and, more recently, Google, alleging unauthorized copying and AI training. Loevy + Loevy, a civil rights law firm involved in the litigation, stated on its website: “If independent voices stay silent, AI companies will lock in a business model that treats creative work as free fuel. If we act now, together, we can force them to license, pay, and respect the artists who make music worth hearing.”
Licensing negotiations between Suno and two of the three major labels collapsed earlier this year. The divide now centers on whether AI-generated songs can be downloaded from the platforms that produce them. Warner Music Group, which settled with Suno in November, permits paid users to download and commercialize their creations. Universal Music Group, which settled with Udio in October, required Udio to disable all downloads and gave users a 48-hour window to export their songs before the feature was removed.
Universal and Sony Music have gone to court to access the terms of Warner’s deal, arguing they cannot negotiate blind against an agreement whose contents they are not permitted to know. Sony, which has refused to settle, added more than 61,000 recordings to its case against Suno. That case is headed for a summary judgment hearing this month. The ruling could determine whether training AI models on copyrighted recordings without a license constitutes fair use or infringement.
Colonial Echoes and Alternative Models
A paper published in April in the Eastern African Journal of Humanities and Social Sciences argues that AI training datasets are “political artifacts that fix neo-colonial forms of oppression by freezing dynamic, living cultures to be consumed exogenously.” It traces the pattern back to Solomon Linda’s 1939 recording of “Mbube,” which was taken from its creator, circulated globally, and monetized by everyone but him. The paper notes the irony that a dataset named Sleeping-DISCO sits at the center of the current controversy.
Legal scholars Chijioke Okorie and Melissa Omino have advanced a related argument about proposed licensing frameworks, observing that standard open licenses treat African dataset owners as if they were well-resourced Global North entities. The result, they argue, is an inequity born of treating unequally situated actors as equals.
Some builders on the continent are responding by rejecting the scraping model entirely. Nigerian start-up Korin AI, for example, works by licensing music from local artists.
