AI training on music datasets without permission

AI training on music datasets without permission has been highlighted by reporting that identified several very large collections of recorded music used to train generative models. Reported music datasets include one collection of 12 million tracks, another of 9 million tracks, and two smaller collections each containing more than 100,000 tracks. The largest collection was reported as large enough that listening to every track would take 91 years, and the song datasets have been downloaded thousands of times.

An organisation called AI Watchdog, launched in September 2025, reported that the training material extended beyond recorded music to include more than 7.5 million books, 81 million research articles, 15 million YouTube videos and writing from tens of thousands of movies and television shows. Reporting linked particular AI music services to the datasets and a news outlet published a searchable database showing music used by those services. Technology companies were also named in reporting as having downloaded music from shared archives for training purposes.

Artists, labels and rights bodies made specific statements about catalogue inclusions and inquiries: one label representative said over 100 songs from their catalogue were used to train AI models, and one artist stated that 138 of their songs appeared across two datasets. Reports noted that Google and Stability AI downloaded material from the Free Music Archive to train models, and a regional rights organisation announced it will launch an investigation into Australian and New Zealand songs, saying ‘No permission, no licence, no payment’ and describing those works as the life’s work of songwriters.

Reported music datasets used to train AI models include one collection of 12 million tracks, a second collection of nine million tracks, and two smaller collections each containing more than 100,000 tracks. The 12-million-track collection was reported as large enough that listening to every track would take 91 years. The song datasets have been downloaded thousands of times. Reporting identified specific catalogue-level inclusions within these collections, including a claim that over 100 songs from one label’s catalogue were used to train AI models and a claim that 138 songs from one artist appeared across two datasets.

An organisation called AI Watchdog, launched in September 2025, reported that the training material extended beyond recorded music to include more than 7.5 million books, 81 million research articles, 15 million YouTube videos and writing from tens of thousands of movies and television shows. Reports named the Free Music Archive as a source of training material and stated that Google downloaded material from the Free Music Archive to train AI models, and that Stability AI has trained on the Free Music Archive. A news outlet published a searchable database showing music used by particular AI music services. The available reporting described the collections as very large and broadly sourced.

Vince Valholla posted on X that over 100 songs from their catalogue were used to train AI models, and Sophia_hjkl said Suno and Udio used 138 of their songs across two datasets, which they described as almost their entire catalogue.

DJ Sabrina The Teen DJ stated that the datasets led to accusations of AI slop and addressed the issue on X, and a news outlet published a searchable database showing extensive usage of artists’ catalogues by Suno and Udio.

For named individuals Emily Swingle, Alex Reisner, Quadeca, Dean Ormston, Nick Cave, and Kylie Minogue, the available sources do not provide this information.

Suno and Udio are identified as AI music services whose datasets were included in reporting, and reports state that Google downloaded material from the Free Music Archive to train AI models while Stability AI has also trained on the Free Music Archive.

The Free Music Archive is named as a source of training material, and the datasets of songs have been downloaded thousands of times.

Genius, YouTube, the Common Crawl Foundation, and Reddit are listed among organisations and platforms mentioned in coverage, but the available sources do not provide specific roles for those entities in the dataset controversy.

AI Watchdog is named in reporting as identifying large-scale inclusion of non-music material in the datasets, and APRA AMCOS announced it will launch an investigation into Australian and New Zealand songs while using the phrase ‘No permission, no licence, no payment’.

MusicTech reported on the story and The Atlantic published a searchable database relating to music used by Suno and Udio.

For other named organisations or contributors mentioned in the coverage, the available sources do not provide this information.

Vince Valholla posted on X that over 100 songs from their catalogue were used to train AI models. He wrote: ‘Late last night I found out over 100+ songs from our catalogue were used to train AI models.’

He also posted: ‘They literally scraped the best songs from our catalogue. I’m sick.’ He added: ‘Until the major labels go through their lawsuits, there’s no way for artists or labels to fight back.’

Sophia_hjkl stated that Suno and Udio used 138 of their songs across two datasets. Sophia_hjkl said: ‘This is almost my entire catalogue of music.’

DJ Sabrina The Teen DJ said the datasets led to accusations of AI slop and addressed the issue on X. DJ Sabrina The Teen DJ wrote: ‘it’s funny how there were no accusations of my music sounding like AI slop until these datasets started getting used to generate slop.’

APRA AMCOS announced it will launch an investigation into Australian and New Zealand songs. APRA AMCOS used the phrase ‘No permission, no licence, no payment.’ APRA AMCOS said: ‘These are not bargaining chips, they are the life’s work of Australian and New Zealand songwriters.’

AI Watchdog, launched in September 2025, reported large-scale inclusion of non-music material in the datasets.

APRA AMCOS announced it will launch an investigation into Australian and New Zealand songs. The organisation stated ‘No permission, no licence, no payment.‘ APRA AMCOS also said: ‘These are not bargaining chips, they are the life’s work of Australian and New Zealand songwriters.‘ The announcement described the action as a launch of an investigation by the rights organisation.

Vince Valholla posted on X that over 100 songs from their catalogue were used to train AI models and wrote: ‘Late last night I found out over 100+ songs from our catalogue were used to train AI models.‘ He also wrote: ‘Until the major labels go through their lawsuits, there’s no way for artists or labels to fight back.‘ Valholla additionally posted: ‘They literally scraped the best songs from our catalogue. I’m sick.‘

AI training on music datasets without permission is the central finding of the referenced investigations and reporting, which identified large aggregated collections of recorded music used to train generative models. The reporting and watchdog analysis documented that those music collections were combined with extensive non-music material, including books, research articles and videos. Artists and labels reported catalogue-level inclusions affecting individual musicians, and rights organisations announced inquiries while some reporting linked companies to downloads from shared music archives used in training. The investigative findings present the unauthorised use of artists’ works in training datasets as the primary issue and note reported constraints on remedies and enforcement available to creators and rights bodies.

Vince Valholla posted on X stating that over 100 songs from their catalogue were used to train AI models and that there was limited recourse until major label litigation proceeded. Sophia_hjkl stated that Suno and Udio used 138 of their songs across two datasets and described this as almost their entire catalogue of music. DJ Sabrina The Teen DJ said on X that the datasets led to accusations of AI-generated material described as ‘AI slop’. For the named individuals Emily Swingle, Alex Reisner, Quadeca, Dean Ormston, Nick Cave and Kylie Minogue, the available reporting does not provide a described role related to the dataset controversy.

Suno and Udio are identified in reporting as AI music services whose datasets were included in coverage of the controversy. The Free Music Archive is named as a source of training material used in the reported collections, and reports state that Google downloaded material from the Free Music Archive to train AI models and that Stability AI has trained on the Free Music Archive. Genius, YouTube, the Common Crawl Foundation and Reddit are listed among organisations and platforms mentioned in coverage, but the available reporting does not specify roles for those entities in relation to the datasets.

AI Watchdog is named as an organisation launched in September 2025 that identified large-scale inclusion of non-music material in the datasets. APRA AMCOS announced it will launch an investigation into Australian and New Zealand songs and used the phrases ‘No permission, no licence, no payment’ and ‘These are not bargaining chips, they are the life’s work of Australian and New Zealand songwriters.’ MusicTech reported on the story, and The Atlantic published a searchable database showing music used by the cited AI music services.

Vince Valholla posted on X reporting that material from their catalogue was included in training datasets and provided multiple statements about that discovery. He wrote: ‘Late last night I found out over 100+ songs from our catalogue were used to train AI models.’ He also wrote: ‘They literally scraped the best songs from our catalogue. I’m sick.’ Valholla added: ‘Until the major labels go through their lawsuits, there’s no way for artists or labels to fight back.’

Sophia_hjkl stated that two datasets included a substantial portion of their work and characterised the impact on their catalogue. Sophia_hjkl said: ‘This is almost my entire catalogue of music.’ The reporting also records Sophia_hjkl’s statement that Suno and Udio used 138 of their songs across two datasets.

DJ Sabrina The Teen DJ addressed the issue on X and commented on public reactions to AI-generated outputs, writing: ‘it’s funny how there were no accusations of my music sounding like AI slop until these datasets started getting used to generate slop.’

APRA AMCOS announced a formal inquiry into affected regional works and used direct language about rights and compensation. APRA AMCOS stated: ‘No permission, no licence, no payment.’ The organisation added: ‘These are not bargaining chips, they are the life’s work of Australian and New Zealand songwriters.’

Reporting also names AI Watchdog as a party that identified wide-ranging inclusion of non-music material in the collections.

APRA AMCOS announced it will launch an investigation into Australian and New Zealand songs in response to reporting on large training datasets. The organisation stated: ‘No permission, no licence, no payment.’ APRA AMCOS also said: ‘These are not bargaining chips, they are the life’s work of Australian and New Zealand songwriters.’ The announcement is described in reporting as the launch of an investigation by the rights organisation.

Affected artists and label representatives issued public statements about catalogue inclusions and legal options. Vince Valholla posted on X: ‘Late last night I found out over 100+ songs from our catalogue were used to train AI models.’ He also wrote: ‘Until the major labels go through their lawsuits, there’s no way for artists or labels to fight back.’ Sophia_hjkl stated that Suno and Udio used 138 of their songs across two datasets and wrote: ‘This is almost my entire catalogue of music.’ DJ Sabrina The Teen DJ commented on X: ‘it’s funny how there were no accusations of my music sounding like AI slop until these datasets started getting used to generate slop.’