Millions of Songs Scraped Without Consent: Australian Artists Find Their Entire Catalogues in AI Training Datasets

A dataset search tool exposed by The Atlantic has confirmed what many musicians long suspected: tens of millions of songs have been scraped from YouTube and used to train AI models without artist consent or compensation. For Australian and New Zealand musicians, the findings are immediate and personal.

Using the tool, artists including Paul Dempsey of Something For Kate and pop veteran Darren Hayes searched for their own work and found it listed in two large music datasets. Dempsey discovered his entire band catalogue alongside his solo recordings. Hayes found his complete 30-year output, including Savage Garden's 1997 hit "Truly Madly Deeply," indexed and ready for machine consumption.

The Datasets in Question

Two datasets sit at the centre of the controversy. The first, Sleeping-DISCO-9M, was assembled by a group calling itself Sleeping AI and contains 9.7 million music tracks sourced from YouTube, paired with lyrics scraped from Genius.com. The second, LAION-DISCO-12M, was compiled by LAION, a Germany-based research organisation, and draws on 12.3 million YouTube tracks. Together, they represent over 22 million recordings harvested from a platform whose terms of service prohibit such bulk extraction.

Other prominent Australian artists confirmed in the datasets include Kylie Minogue, Nick Cave, Powderfinger, and Jimmy Barnes, a spread that signals the scraping was indiscriminate rather than selective.

Industry Response

APRA AMCOS, the Australasian music licensing body representing 128,000 members, has pushed back hard. CEO Dean Ormston accused major technology platforms of bypassing good-faith negotiation entirely, instead lobbying governments and engineering policy proposals specifically designed to eliminate any payment obligation to rights holders. His framing rejects the common industry defence that AI training constitutes a form of fair use or transformative activity.

The response has moved beyond individual organisations. A coalition comprising APRA AMCOS, the Australian Recording Industry Association (ARIA), The Copyright Agency, and the Australian Music Centre jointly issued an open letter calling on lawmakers to strengthen copyright protections in the context of AI training data. The letter reflects a coordinated escalation, not simply reactive frustration.

The broader pattern here mirrors disputes playing out across publishing, visual art, and film: creators discovering after the fact that their life's work was ingested at scale, with no licensing agreement, no notification, and no revenue share. For musicians like Darren Hayes, the dataset search tool has turned an abstract policy debate into a concrete legal and ethical grievance backed by documentary evidence.