The question of who gets paid when an artificial intelligence system learns from someone else’s work has moved from a niche legal argument into one of the defining fights of the creative economy, and Australia’s musicians have decided they can no longer wait for the courts overseas to settle it for them.
That frustration found a louder voice this week after an award win handed one of the sector’s advocates, Partos, a bigger stage to argue that generative AI is being trained on Australian songs, melodies and recordings without consent or payment, and that governments have been slow to respond. As reported by TechDay Australia, the central claim is stark: on current trends, the unchecked use of copyrighted music to build and run AI models could put around AUD $227 million of creator revenue at risk by 2028.
The number behind the alarm
The $227 million figure is the headline of a body of economic research the industry has been citing to put a dollar value on an otherwise abstract risk. It models what happens as AI-generated music, much of it built on the back of existing human work, floods streaming catalogues and licensing markets, eroding the royalties, synchronisation fees and performance income that working musicians rely on. The concern is not simply that a machine can now produce a passable pop track in seconds. It is that the raw material for those machines was, in many cases, the life’s work of people who were never asked and never compensated.
For a sector where most participants are not stadium headliners but session players, songwriters, producers and composers scraping together a living from many small income streams, a figure of that size is not an abstraction. Peak bodies such as APRA AMCOS, which collects and distributes royalties on behalf of tens of thousands of Australian songwriters and publishers, have spent the past two years warning that generative AI represents both an opportunity and an existential threat, depending almost entirely on whether a licensing framework is built around it.
Partos used the award platform to press that same point, arguing that the solution is not to ban the technology but to make its training inputs lawful. In practice that means a system where AI developers must seek permission to use copyrighted material, and must pay for it, in the same way a broadcaster licenses a song or a film-maker clears a sample.
Two ways of seeing the same technology
On one side sit the creators and the collecting societies, who frame the issue as a straightforward property right. Their argument runs that if a company builds a commercial product by ingesting millions of protected works, the people who made those works are entitled to consent and a cut. They point to the speed at which AI music tools have improved and warn that, without rules, the market will simply route income away from humans and towards whoever owns the model.
On the other side are many technology developers and some economists who argue that training an AI on existing material is closer to learning than to copying, and that heavy-handed licensing obligations could lock Australia out of a fast-moving field while doing little to help individual artists. This camp tends to favour broad exceptions for text and data mining, the kind of carve-outs that would let models train on large datasets without negotiating rights one track at a time. They also note that overseas platforms will keep scraping Australian content regardless of what Canberra legislates, which raises the awkward question of whether domestic rules would protect local creators or merely handicap local innovators.
Both positions contain a grain of truth, which is precisely why the policy has been so hard to settle. A licensing market that is too cumbersome could stall legitimate research. A free-for-all could hollow out an industry that contributes billions to the economy and underpins Australia’s cultural identity.
Why this lands hard in Australia
The Australian stakes are sharpened by the structure of the local market. The country produces far more music than it can commercially absorb at home, so its artists depend heavily on export income, streaming payouts and licensing deals struck with global platforms. Those are exactly the revenue lines the research says are most exposed to AI substitution. A loss of $227 million by 2028 would not fall evenly; it would concentrate on the mid-tier and emerging artists who have the least room to absorb it, and who are already contending with streaming economics that pay fractions of a cent per play.
There is also a sovereignty dimension that fits a pattern FluentSea readers have seen across other beats this year, from sovereign AI infrastructure to debates over Australian data being used to train foreign models. The music fight is a cultural version of the same anxiety: valuable Australian material is being absorbed into systems owned and operated offshore, with the economic benefit flowing out of the country unless local rules intervene.
Government is not silent on the question, but it has been deliberate to the point of caution. The Attorney-General’s Department has been running a reference group bringing together rights holders and technology interests to work through copyright and AI, and the Productivity Commission has flagged the broader economic trade-offs of how Australia treats data and intellectual property in the AI era. What has not yet landed is a decision, and that gap between consultation and law is the source of much of the sector’s impatience.
What happens next
The near-term path runs along two tracks. The first is legal, as courts in the United States and Europe work through a wave of cases testing whether training on copyrighted material counts as infringement or fair use. Those rulings will shape the commercial terms that global AI companies offer, and Australian rights holders will inherit the consequences whether or not they are party to the litigation.
The second is domestic policy, where the pressure is now on the federal government to decide whether Australia adopts an opt-in licensing model favoured by creators, a broad data-mining exception favoured by developers, or some negotiated middle ground such as a collective licensing scheme administered through bodies like APRA AMCOS. Each option carries winners and losers, and each will be fought over clause by clause.
For now, the award win has done what awards sometimes do, which is to lend a credible messenger a moment of national attention. Whether that moment converts into the licensing rules Partos is calling for, or simply adds another submission to a growing pile, will depend on how quickly policymakers are willing to move while the $227 million is still a forecast rather than a tally of what has already been lost.
Sources: TechDay Australia
