Australian collecting society APRA AMCOS has accused AI companies of “stealing” music from the likes of Midnight Oil, Sia, Crowded House, Lorde and Yothu Yindi to train generative AI models that will then compete with the exact same artists.
The rights organisation has used a recent investigation by The Atlantic to back up that claim, publishing the music theft allegations as it pushes back at increased lobbying activity by the tech sector in Australia. Lobbying work that aims to secure new copyright loopholes that would allow AI companies to use existing content when training AI models without getting permission from creators or rightsholders
APRA AMCOS CEO Dean Ormston says “AI companies are asking the Australian and New Zealand governments for a copyright carve-out”, in part by promising to make investments in the two countries in return. But, as that lobbying continues, “we can show you exactly what they have already taken” with “no permission, no licence, no payment”. The many songs and recordings that have been “taken”, he adds, “are not bargaining chips, they are the life’s work of Australian and New Zealand songwriters”.
In an accompanying statement, the collecting society explains that “a new investigation by The Atlantic has uncovered millions of songs, including Australian and New Zealand musical works, that have been taken as part of four giant datasets of songs that are being shared within the AI-development community”.
A data tool from The Atlantic, it goes on, “allows anyone to look up an artist and see exactly which of their tracks have been fed into AI training systems without consent, without a licence and without payment”.
Name-checking many more Australian artists included in those four giant datasets - including INXS, Kylie Minogue, Nick Cave, Tame Impala, Bic Runga, Fat Freddy’s Drop and Cat Empire - APRA AMCOS argues that “the songs that built Australian and New Zealand culture” have been exploited in AI training without the legally required permission while “tech companies lobby to trade investment for immunity”.
Most AI companies are very vague about what existing content they have used to train their models, much to the annoyance of the creative and copyright industries, including the music industry, which want transparency obligations to be forced onto AI businesses via new laws.
Given the lack of transparency over AI training, The Atlantic’s investigation is interesting. Although, because we know both Udio and Suno were basically trained by scraping all the music off YouTube, the industry has been more or less operating on the assumption for a while now that pretty much all commercially released music has been used by at least some AI companies to train generative AI models.
APRA AMCOS using The Atlantic’s report in this way at this time is a sign that the creative and copyright industries are increasingly concerned about the intensive lobbying being employed by the tech sector as it seeks to influence an Australian government that last year said it would not be introducing any new copyright exceptions to benefit AI companies.
Earlier this month Scott Farquhar, Chair of the Tech Council Of Australia, strongly criticised the country’s copyright regime. He said, “if I train in Australia, I need to cut a deal with every single recording artist in the entire world, because of the way our copyright laws work”. Reckoning that simply can’t be done, he insisted that, without copyright reform, building music AI platforms in Australia is “impossible”.
Australian record industry trade group ARIA was quick to dub that claim “nonsense”, with CEO Annabelle Herd stating, “You could license around 80% of the world’s sound recordings to train AI globally - including in Australia - with four deals: one with each major record label and one with Merlin”.
And given Australia is “home to one the world’s great creative and media cultures”, she added, “we will not rewrite its copyright laws on the advice of the people who stand to profit most from dismantling them”.
In his new statement this week, APRA AMCOS boss Ormston says that - since the Australian government confirmed it would not introduce any new copyright exceptions for AI companies, meaning they would need to negotiate licensing deals - “major tech platforms have not come to the table - not once”.
“Instead”, he adds, “they have lobbied governments, circulated policy papers and proposed solutions designed to extinguish any obligation to pay. The only path forward is a genuine licensing conversation with the people whose work they have been using. We are ready. We have always been ready. The question is whether they are”.