Artificial intelligence can now create a song in less time than it takes the music industry to work out who should be paid for one. That gap may prove to be one of the defining challenges of the next phase of digital music. Much of the debate around generative AI has focused on the most immediate questions. Was copyrighted music used to train a model without permission? Can an artist’s voice be replicated? Will AI generated songs compete with human musicians?Â
Generative AI does not simply make it easier to produce music. It could dramatically increase the number of ways in which music is created, adapted, personalised and reused. Every one of those uses potentially carries rights, ownership and payment implications. The music industry therefore faces two distinct conversations around AI. The first concerns AI technology used to create music or train models on copyrighted works, raising questions around legislation, licensing and how protected works can be used. The second concerns how AI technology can be used within Collective Management Organisations (CMOs) themselves to improve and accelerate business operations, including the systems responsible for identifying rights and distributing royalties.
The real question is not whether AI will become part of music. The question is whether the rights infrastructure around music can evolve quickly enough to keep pace.
Music Has Been Here Before
There is a temptation to view generative AI as an entirely unprecedented disruption, but the music industry has experienced this pattern before. When Napster emerged at the end of the 1990s, it revealed something important about consumer behaviour. People wanted immediate access to enormous catalogues of music. The problem was that the service providing that experience had collided with copyright and the existing commercial model. What followed was years of conflict, experimentation and eventually adaptation. Digital music did not disappear. The behaviour that Napster helped popularise became part of a licensed market through downloads and then streaming.
That history offers a useful way of thinking about generative AI. The current dispute should not necessarily be interpreted as evidence that the technology and the music industry are fundamentally incompatible. It may instead represent the difficult opening stage of another transition towards a commercial model that has not yet fully formed. At the same time, copyright questions remain actively contested. In July 2026, Sony Music Entertainment filed a new lawsuit against Udio alleging that more than 30,000 Sony-owned recordings had been copied for use in training its models.
But licensing is only one side of the transformation. As consumption moved from physical sales and downloads towards billions of digital interactions, Collective Management Organisations (CMOs) and other rights organisations had to manage growing volumes of repertoire, usage and ownership data and connect those uses with the creators and rights holders who should be paid.
AI represents another potentially transformative phase for music, and the industry again has to consider both the rules governing new technology and the operational systems required to support the market that develops around it. Questions about whether copyrighted works can be used to train generative AI, and under what licensing framework, will ultimately require answers from rights holders, policymakers and the wider industry. Alongside that debate is a different technological challenge for CMOs: how AI can be used within their own operations to manage increasingly complex rights data, improve matching and reconciliation, and help ensure royalties reach the correct creators and rights holders efficiently.
The Bigger AI Problem May Be Getting People Paid
Music rights are already complicated before artificial intelligence enters the picture. A single recording can involve rights in the composition and the sound recording, several songwriters and performers, publishers, labels and collecting organisations. Ownership can vary between territories and rights can change hands over time. The underlying data is equally complex. The title of a recording may appear differently in separate databases, contributors can be listed in inconsistent ways, recordings must be linked to the correct compositions and duplicate or conflicting claims can emerge between different sources.
The industry has developed sophisticated systems for managing this information, but the fundamental challenge becomes harder as the volume of music usage grows. This is particularly visible in the long tail of royalties. Some transactions are worth enough to justify detailed investigation when information is incomplete or uncertain. Others are extremely small. In those cases, the cost of having a person investigate an uncertain match can exceed the royalty that might eventually be distributed. That creates an uncomfortable economic reality in which a creator can be entitled to money while the cost of identifying and processing that entitlement makes individual investigation impractical.
Generative AI could magnify this issue considerably. The Mechanical Licensing Collective (The MLC) reported that it processed $943.1 million in royalties during 2025, with an average initial match rate of 85%. Even after continued reprocessing improved match rates, approximately $328.2 million in royalties remained unmatched as of February 2026 because reported sound recording uses had not yet been successfully linked to registered musical works. Each interaction could generate information that has to be connected back to the relevant works and rights holders. The challenge is therefore not simply that more music can be produced. It is that the number of uses requiring identification, attribution and potentially payment could grow at a rate that current levels of automation in rights administration cannot realistically match.
Licensing therefore solves only one part of the problem. The advent of streaming already transformed the scale of rights administration by creating an enormous increase in usage data. A radio station might play between 250 and 350 songs in a given day, while a major streaming service can record billions of streams across millions of different tracks. On 24 December 2025 alone, Spotify recorded more than 11 billion streams globally.
AI-generated music could introduce another significant increase in complexity. A single AI-generated song may draw on hundreds or potentially thousands of underlying musical works from the corpus used to train the generation tool. Instead of administering rights between perhaps two and ten copyright holders associated with a musical work, the industry could potentially need to account for hundreds or thousands of rights holders, each with a very small attributable share.
Emerging attribution technologies may help identify which underlying works contributed to a specific AI-generated song, but identification solves only part of the problem. Rights systems must then determine who owns those works, the relevant shares and territories, and how potentially enormous volumes of very small percentage interests should be administered and paid. At sufficient volume, those questions become as much a data and infrastructure challenge as a legal one. If music creation and reuse can happen at machine scale, rights identification, licensing and payment will increasingly need to operate at machine scale too.
AI Can Help Solve the Problem It Creates
AI is usually presented as the source of the new challenge facing music rights, but it can also be part of the solution. Large music rights datasets contain exactly the type of problems that computational systems are increasingly capable of helping with.
AI can help identify likely relationships between recordings and compositions, examine inconsistent information across datasets, surface possible duplicates and identify cases where ownership information appears to conflict. Rather than requiring people to inspect every transaction, technology can narrow the problem and direct the most difficult cases towards human review. That changes the economics of rights management because the most interesting use of AI is not necessarily doing something that a person already does more quickly. It is making possible work that was previously uneconomic to perform at all.
Consider thousands of low-value royalty transactions that contain some uncertainty. Employing people to manually investigate every case may make no commercial sense. A system capable of examining those transactions automatically and escalating only the uncertain cases changes the calculation. Music that might previously have remained unmatched can potentially be connected with the correct work, while payments that were too costly to investigate individually can become economically viable to process.
There is an important limitation though. AI systems can produce answers that sound convincing and are completely wrong. That is particularly dangerous where intellectual property and payments are involved. For this reason, AI should not simply be given a dataset and trusted to make final decisions. The strongest applications combine AI with conventional software and clearly defined controls. AI might propose that two pieces of information refer to the same musical work, but the surrounding system should then test that conclusion against identifiers, ownership data and defined rules before acting on it.
This approach recognises what AI is good at while protecting against what it is not. Models are powerful at finding patterns and interpreting complex information. They are not inherently reliable arbiters of whether a financial or rights decision is correct. Important systems need ways to test whether their outputs make sense.
Rights Infrastructure Will Decide What Comes Next
The extraordinary attention being given to individual AI models can distract from where much of the long-term value will actually be created.
That layer contains the trusted rights data, ownership information, permissions, business rules and validation processes that turn a general AI capability into something useful for the music industry. Platforms such as Matching Engine sit within this infrastructure layer, bringing together repertoire, ownership and usage data so that rights information can be matched, validated and processed within the wider royalty administration workflow. A model cannot inherently know which songwriter owns a particular percentage of a composition or what agreement governs a use in a specific territory. It needs reliable information and systems around it.
The discussion today is understandably dominated by what AI companies are allowed to do with music. The discussion tomorrow is likely to focus increasingly on how rights holders can permit new uses while still identifying them and getting paid. Copyright can establish the rules of the market, but infrastructure has to make those rules work. A licensing agreement is only valuable if the industry can recognise when licensed material has been used and accurately distribute the resulting value.
That could become considerably harder in a world of personalised music. Instead of one recording generating millions of streams, we may begin to see millions of individual musical experiences generated from licensed material. The defensive instinct is therefore understandable, but the opportunity is larger than defence. Artists and songwriters could choose which forms of generative use they permit. Fans could participate more directly in music while creators share in the value produced. Existing catalogues could generate types of engagement that simply did not exist in the traditional recording or streaming model.
For that future to work, the infrastructure used to administer music rights must continue to scale with the volume and complexity of music usage.. That means systems capable of managing repertoire and ownership data, matching recordings to compositions, resolving conflicting or duplicate claims, applying the relevant rights and royalty rules, and processing distributions accurately at scale. The music industry has already learned what happens when consumer technology develops faster than its commercial infrastructure. The answer was eventually not to stop digital distribution but to build systems capable of licensing and administering it at scale. Artificial intelligence presents a new version of that challenge. The technologies needed to support this transition are already available. The priority now is for music organisations to adopt and integrate proven rights management technologies that can help them manage increasing data volumes and complexity while adapting to an evolving regulatory and commercial landscape.
The race is not simply between AI and musicians. It is between the speed at which AI can create new ways of using music and the speed at which the industry can identify those uses, license them and ensure that creators are compensated. If rights infrastructure can keep pace, generative AI does not have to weaken the economic relationship between technology and creativity. It could ultimately make that relationship stronger.
