Prison for the first AI music streaming fraud: US prosecutors dismantle the 'invisible harm' defence

By Zendoric · September 29, 2026.

Michael Smith, 54, pleaded guilty in March to conspiracy to commit wire fraud (the offense of using remote communications -here, streaming platforms- to defraud) after admitting that, for more than six years and since 2017, he used artificial intelligence to generate thousands of songs and play them billions of times through bot networks on Spotify, Apple Music and other services. He agreed to return more than $8 million obtained from that fraud, according to his own plea agreement.

Now prosecutors are asking for real prison time. In a filing submitted on Tuesday, September 29, prosecutors asked the judge for at least 46 months behind bars -the bottom of the federal guidelines range for his charge, which runs from 46 to 57 months and carries a statutory maximum of five years-. It is a notably tougher request than the recommendation from the federal probation office, which had proposed only 24 months. Last week, Smith's lawyers had asked for exactly the opposite: zero days in jail.

The clash is not only about the sentence, but about what 'causing harm' means. Smith's defense argued that the damage to any specific artist was 'insignificant,' a 'drop in an ocean' spread across millions of songwriters and artists, and that 'no artist suffered perceptible harm.' Prosecutors respond with the real mechanics of the business: streaming royalties are paid out of a finite revenue pool, so each of Smith's fraudulent plays was not neutral -it diluted the value of every legitimate play by every other artist-. In their words, the money Smith diverted came directly from the creators whose legitimate royalties were taken.

That argument is what makes this the first criminal enforcement case in the United States for streaming fraud, a problem the industry has been dragging along for years -the IFPI, the federation representing record labels globally, has launched specific initiatives to fight it- and one that generative AI has made considerably worse: generating thousands of filler tracks costs practically nothing today, and automating their playback with bots is trivial. Before, setting up fraud on this scale required buying or stealing real music; now it is enough to ask a model for it.

Our reading: the defense's argument -that the harm dilutes until it disappears- is exactly the kind of alibi that a misunderstood generative abundance puts on the table, and that is why this case matters beyond music. AI can generate real abundance -content, diagnoses, code- when it creates new value; but when it is used to artificially inflate a finite pool of resources shared among many (royalties, ad clicks, reviews, traffic), it generates nothing: it merely redistributes other people's value toward whoever games the system. These are two uses of the same technology with opposite signs, and confusing them -treating parasitic extraction as if it were creation- is one of the real short-term risks this wave of cases is beginning to expose.

The precedent also matters for what comes next. If AI streaming fraud is treated as serious white-collar crime -with sentences measured in years, not months of probation- it raises the stakes for anyone thinking of automating similar fraud in other markets with a fixed reward pool: programmatic advertising, verified reviews, or future revenue-sharing systems among AI agents. If instead the 'distributed harm, therefore invisible' argument prevails, that logic will be left open to replicate in any sector where ownership is spread among many. The judge who sets Smith's sentence is not only deciding months in prison: he is setting how much it costs, in the United States, to use AI to steal from a crowd.