Kylie Minogue has become the latest high-profile artist drawn into the expanding debate over how artificial intelligence systems are trained, after an industry group said AI models have already learned from her songs without payment. The same claim applies to Nick Cave, another globally recognized Australian musician. The allegation is not that fans are listening to copies of their music, but that machine-learning systems ingested recordings and compositions as training material and then built commercial products on top of that knowledge without licensing the work or sharing revenue with the creators.

The dispute has moved quickly from a niche copyright argument to a national policy fight in Australia. The International Confederation of Music Publishers says AI companies have asked dozens of governments to loosen copyright rules so that models can be trained on protected material more easily. Most of those governments have apparently resisted. Australia, however, is reportedly the only country giving the request a second look, putting the country at the center of a global test over whether creative work can be used as raw material for AI.

Key facts at a glance

• Kylie Minogue and Nick Cave are named as artists whose songs have been used to train AI models without compensation, according to the music publishers’ group.
• The International Confederation of Music Publishers says AI firms have lobbied dozens of governments to relax copyright rules for AI training.
• Australia is described as the lone country among those approached that is reconsidering the request.
• The group’s chief executive accused AI companies of asking lawmakers to ‘legitimise larceny’ and of trying to strong-arm Australian policymakers with investment threats.
• Prime Minister Anthony Albanese said in July that it amounts to theft if creators cannot decide whether AI learns from their work or how much they earn from it.
• A parliamentary AI committee has revived the issue, with lawmakers from both sides calling the situation a deadlock.
• Tech companies argue that the most advanced AI models will not be trained in Australia unless local copyright laws change.
• If Australia gives in, it would be the odd one out among the countries tech firms approached.

Why Minogue and Cave are central examples

Minogue and Cave are not random names in a policy paper. Both are Australian cultural exports with catalogs that stretch across decades and formats. Minogue first became a household name through television before building a pop career that has produced dance-floor staples, arena tours and a durable global audience. Her catalog includes eras that span synth-pop, disco, electronic music and contemporary pop, and songs such as ‘Can’t Get You Out of My Head’ and ‘Padam Padam’ have become reference points far beyond Australia. Cave’s work, first with The Birthday Party and later with Nick Cave and the Bad Seeds, has moved through post-punk, gothic rock, ballads and experimental songwriting. His albums and live performances have made him one of the most distinctive voices in Australian music.

That matters because AI training depends on volume, variety and cultural relevance. A model does not need to reproduce an entire song to learn patterns from it. It can ingest lyrics, melodies, rhythms, production styles and vocal characteristics across thousands of recordings. The resulting system may then generate new material that sounds like a genre, a period or even a particular artist. For rights holders, the concern is not only that a copy might appear. It is that the model itself can become a commercial substitute for the labor and investment that created the original work.

The allegation: training without permission or payment

The music publishers’ group says the use of Minogue’s and Cave’s songs is part of a wider pattern in which AI systems have learned from copyrighted music without the artists seeing a cent. That claim goes to the heart of the debate. AI developers often describe training as a form of reading or analysis, arguing that the model learns statistical relationships rather than storing copies. Rights holders reply that the process still requires copies of protected works, and that the commercial value of the resulting model is built partly on those works. They want consent, transparency and compensation, not merely a promise that outputs will be filtered later.

The group’s chief executive did not soften the language. He accused AI firms of asking lawmakers to ‘legitimise larceny’. He also said they were trying to strong-arm Australian policymakers with a message that if Australia does not change its rules, they will not invest. That accusation turns a technical copyright question into a question of bargaining power. The companies frame relaxed rules as a condition for innovation and investment. Critics frame the same request as a demand that the law treat uncompensated use of creative work as normal.

The political fight in Australia

The Australian government has been under pressure from both sides. In July, Prime Minister Anthony Albanese took a firm public stance, saying it would be theft if creators could not decide whether AI learns from their work or how much they earn from it. That language suggested the government was prepared to defend creator rights. Months later, the issue has returned through a parliamentary AI committee, where lawmakers from both sides have described the state of play as a deadlock. The committee’s renewed attention shows that the lobbying campaign has not gone away.

The argument now being aired is that AI models in the United States already learn from copyrighted material without creators getting paid. The companies building the most advanced models, the argument continues, will not train them in Australia unless local copyright laws change. That creates a familiar policy threat: change the rules or lose the investment. For creative industries, the fear is that this threat will be enough to produce a broad exception that weakens licensing markets and makes it harder for artists to negotiate.

Copyright, licensing and the missing royalty check

Music is not a single right. A recording involves performers, producers and sound-recording owners. A song involves songwriters and publishers. A lyric may have its own protections. AI training can touch all of these layers at once. A licensing system that works for one layer may not work for another. That complexity is one reason rights holders want clear rules rather than a general exception. They argue that collective licensing, disclosure requirements and record-keeping obligations could allow AI development while ensuring that creators are paid.

The alternative, they warn, is a royalty-free pipeline. If AI companies can train on entire catalogs without permission, the market for licensing music as training data may never develop. Artists would have little ability to opt out if they do not know what has been ingested. Publishers would have little leverage if the law says the ingestion is permitted. Smaller creators, who cannot afford long litigation, would be especially exposed. The result could be a system in which the most valuable cultural works feed the next generation of commercial AI products while the people who made those works receive nothing.

The investment argument and its limits

AI companies have a genuine interest in predictable rules. Training large models is expensive, and legal uncertainty can deter investment. But the investment argument cuts both ways. Australia’s creative industries are also an economic asset. Music, television, film, games and publishing employ thousands of people and export Australian culture around the world. A copyright framework that weakens creator rights may attract AI firms in the short term, but it can also undermine the industries that supply the raw material. The question is not whether Australia wants AI investment. It is what price the country is willing to pay for it.

There is also a practical problem with the threat that companies will train elsewhere. If models are trained in countries with looser rules and then used in Australia, Australian law may still have little control over the training process. That does not make domestic rules irrelevant. It means Australia has to decide what it will permit on its own soil, what transparency it will require from models sold to Australians, and whether it will support licensing systems that can operate across borders. A race to the bottom on copyright would not necessarily protect Australian creators, and it might not even secure the investment the companies are promising.

What the committee deadlock means

The parliamentary committee deadlock reflects a broader split. One side emphasizes competitiveness, innovation and the risk of falling behind in AI. The other emphasizes consent, compensation and the cultural value of creative work. Both sides can point to real concerns. AI development can bring productivity gains and new tools for artists. It can also produce models that imitate voices, styles and entire genres at scale. The policy challenge is to capture the benefits without treating creators as a free resource.

For Minogue and Cave, the immediate issue is not whether AI can make music. It is whether the laws governing that music recognize a right to consent and payment. The same question applies to thousands of smaller Australian artists who do not have global fame but whose work may still be swept into training datasets. The committee’s next move will determine whether Australia treats those rights as a foundation for innovation or as an obstacle to be bargained away.

Source: Digital Trends News