Imagine playing a packed room where nobody has ears. The seats are occupied, the attendance looks terrific, and the money arrives on time. There is just one small problem: the audience is software. That is the part of this week’s AI music fraud story worth sitting with, once the novelty of computer-generated songs wears off.
On October 6, Michael Smith of North Carolina was sentenced to 18 months in prison after pleading guilty in March to conspiracy to commit wire fraud. According to the U.S. Justice Department’s sentencing announcement, his scheme used automated accounts to stream music he controlled and collected more than US$8 million in royalties. The court also ordered just over US$8.09 million in forfeiture.
AI supplied an enormous catalogue. Bots supplied the supposed listeners. It is tempting to bundle those ingredients into a single verdict about the future of music, but they answer different questions. How a recording was made matters. Whether anyone actually chose to listen matters too, and here it was the invented audience that made the royalty claim fraudulent.
That distinction gives this story a longer shelf life than another argument about whether a machine can write a decent chorus. A terrible song can have genuine fans. A beautifully performed song can have fraudulent streams. Taste is a wonderfully messy human business; checking whether the audience exists ought to be a more basic requirement.
The money follows the listening
Streaming can feel like a digital jukebox: press play, send a tiny coin to the artist. The accounting is less direct. In its royalties guide, Spotify explains that it pays rightsholders according to their share of streams, rather than a fixed fee for each play. Those rightsholders then pay artists and songwriters according to their agreements.
That makes the integrity of the count important. In a shared royalty pool, fabricated listening can redirect money that would otherwise be allocated to legitimate streams. The harm does not require an actual listener to be fooled into loving a fake band. It can happen behind a perfectly ordinary listening session, while someone else manipulates the numbers used to divide the money.
Smith’s operation ran from 2017 to 2024. In WUNC’s October 7 reporting, the expansion into AI music around 2018 is a striking detail: a musician with a small catalogue gained access to an industrial volume of songs. The technology helped provide the scale for the operation. It did not provide a community of people who cared about those recordings.
This is where the familiar language of creator success becomes slippery. A stream count is useful because we assume it stands for something beyond itself: somebody heard a song, perhaps enjoyed it, perhaps came back. Once the count becomes the product, that connection can disappear. The dashboard still gets busier. The relationship it is supposed to describe never happens.
For a working musician, that is a particularly grim kind of competition. Writing, rehearsing, recording and persuading strangers to give you four minutes are difficult enough. Competing against fabricated listening adds an opponent who does not need a good song, a memorable gig or even a bored teenager willing to leave an album running.
Better policing, without a musical purity test
There is an obvious temptation to make AI detection the centre of the response. Labelling synthetic music can help listeners make informed choices, and questions about consent, credit and training material deserve their own serious arguments. But an AI label cannot tell us whether the streams attached to a track came from real listeners. These are separate checks, even when the same operation abuses both.
Spotify’s published artificial-streaming policy says detected artificial plays do not earn royalties, count towards public stream numbers or charts, or positively influence recommendations. It also warns artists about paid services promising streams or playlist placements. Those are useful commitments. They are statements of policy, though, not evidence that every fraudulent play is caught.
The awkward part is making enforcement understandable to the people whose livelihoods depend on it. Spotify acknowledges that artists can sometimes see artificial-streaming spikes in their private dashboards even after associated royalties have been withheld and public figures adjusted. A number on a screen, in other words, is not necessarily a settled accounting of what happened.
That is why transparency should be part of the product, rather than something musicians have to hunt down after a frightening email. Artists need clear explanations when payments or tracks are affected, a practical way to challenge mistakes, and information that helps them distinguish a genuine new audience from suspicious activity. Protecting the royalty pool and treating small artists fairly should reinforce each other.
Listeners need a little humility about numbers, too. Popularity can be a useful invitation to try something; it should not become a substitute for deciding whether we like it. A modest audience can be real and devoted. An impressive counter cannot tell us what a song meant to anybody, even when every play behind it is legitimate.
The sentencing offers accountability for one operation. It does not settle the creative possibilities of AI music, or prove that streaming has solved fraud. What it does offer is a useful question to keep asking whenever a platform celebrates engagement: what human activity is this number supposed to represent?
Music can survive an awful lot of questionable choruses. A system that pays for listening needs to be rather less relaxed about whether listening happened. Somewhere, a musician is trying to reach a person who might carry a song around for years. The counter should be able to tell the difference between that possibility and an empty chair wearing headphones.
