There is a database of 12 million songs. It does not contain the actual music files. It contains directions: links, metadata, and instructions on where to download each track. That distinction matters enormously in court. Because simply listing where music can be found is not, by itself, a copyright violation. The infringement happens when an AI company follows those directions and trains its system on what it finds.
This is the gap that AI music companies have operated inside for the past three years. Build the product, grow fast, and argue in court later about whether any of it was actually illegal.
Suno is the clearest example. The company trained its AI on copyrighted songs, built a platform that now produces roughly seven million songs every day, and reached a valuation of $2.45 billion. Warner Music settled with Suno in late 2025. Universal Music and Sony have not. Their case is heading toward a key court hearing in the United States this month, and the number of songs in the complaint has grown from 560 when it was filed to over 61,000 after audio fingerprinting revealed how much copyrighted material was actually in Suno's training data.
The outcome of that case matters beyond music. If the court rules that training on copyrighted material is legal as long as the output is sufficiently different from the input, the same logic applies to every AI system trained on books, images, or any other creative work. If it goes the other way, every AI company that built on unlicensed material faces a reckoning.
For individual artists, especially those outside the major label system, the picture is already clear: the settlements reached so far do not cover them. Independent musicians are pursuing separate class action lawsuits, but those cases are slow and expensive, and the AI companies already have their valuations.
A global study from the music rights body CISAC found that musicians worldwide could lose around 24% of their total revenues to AI substitution by 2028, amounting to roughly €10 billion in cumulative losses across the music sector alone. That is not a projection about what AI might do. It is a projection about what AI is already doing, as AI-generated tracks flood streaming platforms. In early 2026, the platform Deezer reported receiving around 75,000 AI-generated tracks every single day, representing 44% of all new music delivered to it.
Australia sits at the center of this debate for two reasons. First, Australian music is heavily represented in the training datasets: Kylie Minogue alone has 182 songs in one database. Second, the Australian government almost made it legal. In August 2025, its Productivity Commission proposed exempting AI companies from copyright law when mining data for training. The creative industry responded with force. Artists marched on parliament. Bodies representing over a million workers in Australian creative industries signed open letters. The proposal was killed in October 2025.
But the pressure has not stopped. Tech industry lobbyists are still pushing, and independent senator David Pocock raised concerns in June 2026 that new carve-outs were being discussed inside government. Industry groups estimate that unregulated AI music generation could strip around $519 million from Australian music revenues by 2028.
The more durable fix being watched globally is the EU's approach. From August 2026, any AI model accessed within Europe must publish a summary of what it was trained on, comply with EU copyright law regardless of where the company is based, and allow rights holders to opt out of having their work used. Fines for non-compliance can reach 3% of a company's global annual revenue. That enforcement mechanism is what makes the EU rule meaningful: it applies to a US company the same way it applies to a European one.
Around 300 commercial AI licensing deals have already been signed globally between AI companies and content owners. The music industry's own rights bodies say they are ready to negotiate more. The obstacle is that AI companies have had little incentive to start those conversations while they could grow for free.
For anyone who runs a business that uses music, commissions creative work, or pays for content of any kind, the next 12 months will likely settle whether the cost of AI-generated material goes up sharply or stays near zero. The July 2026 court ruling in the US and the EU enforcement deadline in August are the two dates worth watching.