Ask almost any working Australian musician how they feel about artificial intelligence hoovering up their back catalogue to train song-generating models, and the answer tends to be blunt. They hate it. What they don’t have, according to a recent analysis published in The Conversation, is much in the way of legal protection to do anything about it.
The tension has been building for a couple of years, but it has sharpened as consumer tools such as Suno and Udio have made it trivially easy to type a prompt and receive a fully produced track in seconds. Those systems learned to do that by being trained on enormous libraries of existing recordings — and there is a live global debate about whether that training material was licensed, scraped, or simply taken. For Australian artists, the frustration is compounded by a stark reality: the country’s copyright regime was not built for a world in which a machine can absorb a lifetime of creative output and reconstitute it into something new.
Why the law leaves artists exposed
Australian copyright law protects specific things: the musical work, the lyrics, the sound recording. It gives creators the exclusive right to reproduce and communicate those works. On paper, that sounds like solid armour. In practice, generative AI slips through the gaps.
The first problem is proof. To bring a copyright claim, an artist generally has to show that a substantial part of their particular work was copied. When a model has been trained on millions of songs and outputs a track that sounds like it belongs to a genre or a scene rather than any single recording, it is extraordinarily difficult to point to the specific bar of music that was lifted. The harm feels obvious to the artist; the legal fingerprint is faint.
The second problem is that Australia, unlike the United States, does not have a broad “fair use” defence — it has a narrower set of “fair dealing” exceptions for purposes such as research, criticism and parody. That difference cuts both ways. It means AI developers cannot lean on the flexible American doctrine that is currently being fought over in US courtrooms, but it also means Australia has no settled framework for deciding whether training a model on copyrighted music is lawful at all. The uncertainty benefits whoever has the deeper pockets, and that is rarely the musician.
Then there is the question of what the machine produces. A voice, a style, a vibe — none of these are neatly protected by copyright. An AI track that mimics the timbre of a well-known Australian vocalist without reproducing any specific recording may sail past copyright entirely, even as it trades on decades of an artist’s reputation.
Two sides of a hard argument
The music sector’s position is that consent, credit and compensation are non-negotiable. Peak body APRA AMCOS, which collects royalties on behalf of songwriters and publishers, has warned that unlicensed AI training threatens the livelihoods of the very creators who supply the raw material. Its research has repeatedly pointed to the risk that AI-generated music could siphon income away from human artists in streaming, sync and background-music markets — the bread-and-butter revenue that keeps mid-tier careers alive. The industry line is not anti-technology so much as anti-freeloading: use our work, by all means, but pay for a licence and get permission first.
The technology camp frames it differently. Developers argue that models learn from culture the way humans do — by listening, absorbing and being influenced — and that treating every act of training as an infringement would strangle a promising industry before it matures. They point to the productivity gains, the new creative tools placed in the hands of amateurs, and the risk that heavy-handed local rules simply push AI development offshore while Australians miss out on the upside. Some also argue that clear licensing markets, once established, could become a new revenue stream for rights holders rather than a threat.
Both positions contain truth, which is precisely why the policy question is so fraught. The disagreement is not really about whether AI is here to stay — it plainly is — but about who captures the value it creates, and whether the humans whose work made it possible are cut in.
What it means for Australia
This is a global fight, but it lands with particular force here. Australia has a proud, export-heavy music culture — think of the artists who have carried the country’s sound onto world stages — yet it is a comparatively small market with limited leverage over the multinational tech firms building these models. Local creators cannot count on the sheer scale that gives American or European rights holders a seat at the negotiating table.
There is also a policy vacuum to fill. The federal government has been consulting on the responsible use of AI, and the Attorney-General’s Department has run reviews touching on copyright and AI, but no dedicated legislative fix has landed. The question of whether Australia adopts a text-and-data-mining exception, a mandatory licensing scheme, or a transparency obligation forcing developers to disclose their training data remains open. Whatever Canberra decides will ripple through the livelihoods of thousands of songwriters, session players and producers, many of whom already piece together an income from small, precarious sources.
For emerging Australian artists, the stakes are especially personal. A young musician uploading demos to build an audience may be unknowingly feeding the very systems that could one day undercut them. Without clearer rules on consent and disclosure, the incentive to create and share risks being quietly eroded.
What’s next
Expect the pressure for reform to keep mounting. Industry groups are lobbying for a licensing framework that treats AI training as a paid, permission-based activity, and for transparency requirements that would let artists find out whether their work has been used. Internationally, court decisions in the US and regulatory moves in the European Union — where transparency obligations for AI training data are already being written into law — will shape what Australia feels able to demand.
In the meantime, many artists are taking matters into their own hands: opting out where platforms allow it, adding contractual clauses, and backing collective action through their peak bodies. But opt-outs only work if developers honour them, and enforcement across borders is thin. Until the law catches up, Australian musicians are likely to keep doing what the headline says — hating the way AI uses their songs, while holding very few cards to stop it.
Sources: The Conversation.
