AI Music Streaming Fraud: $8 Million, Billions of Bot Plays, and Prison

Hundreds of thousands of AI-generated songs. Billions of streams. Up to 10,000 fake listeners at once. More than $8 million in fraudulent royalties.

On October 6, 2026, a federal judge in Manhattan sentenced Michael Smith, 54, to 18 months in prison, followed by two years of supervised release. He was also ordered to forfeit $8,091,843.64. Smith had pleaded guilty in March to conspiracy to commit wire fraud. The US Department of Justice confirmed the sentence.

What makes this case extraordinary is not simply that someone used AI to make music. It is that AI gave an existing streaming-fraud operation the one thing it needed to scale: an enormous supply of tracks to play.

Smith was not convicted for making AI music. He was convicted for using fabricated listeners to manipulate royalty payments. That distinction matters.

The verdict at a glance

These figures come from the sentencing announcement and the March guilty-plea announcement.

How did he make millions from songs nobody needed to hear?

According to the original federal indictment, the operation combined five elements:

1. Accounts pretending to be real subscribers. Fake identities, email accounts, payment cards and paid plans gave automated accounts access to streaming services.
2. Cloud-hosted playback. Virtual computers opened streaming players and ran multiple accounts at the same time. Software macros kept the playback going.
3. A catalog large enough to hide the pattern. Instead of playing one track an implausible number of times, the operation distributed plays across many tracks.
4. Mass-produced music. Smith obtained hundreds of thousands of AI-generated songs through collaborators and uploaded them under invented song and artist names.
5. Royalty collection. Streams presented as genuine listening activity generated payments to rights holders controlled by the scheme.

The scale was not an accidental side effect. In a December 2018 email quoted in the 2024 DOJ announcement, Smith said he needed "a TON of songs fast" to get around anti-fraud checks. Generating more audio was a way to distribute the suspicious listening activity across more titles.

The indictment describes randomly generated titles and artist names, including tracks called "Zygophyceae" and "Zymogenic" and an artist name "Calm Baseball." A song did not need to become popular. It needed to look like another ordinary entry in a huge catalog.

The early calculation: 661,440 streams per day

The indictment cites an October 2017 email in which Smith laid out his own model:

That is approximately $3,307 per day in his spreadsheet-style projection. It was Smith's estimate in an email, not an independently verified daily payout or a universal streaming rate. The indictment, page 7, records the assumptions.

The later operation was larger: the 2026 DOJ statement says Smith sometimes ran as many as 10,000 bot accounts at once.

The Taylor Swift comparison is real — but very specific

One number from the sentencing announcement is almost absurd:

In this specific slice of YouTube Music family-plan activity, Smith's bots generated roughly 8.7 times as many streams as Swift's catalog.

This does not mean Smith was more popular than Taylor Swift across YouTube, Spotify or the music industry. It is a comparison restricted to one service, one month and one subscription category. The plays attributed to Smith were fraudulent, not evidence of an audience. The figures are taken directly from the DOJ's October 6 statement.

That qualification matters. The real story is disturbing enough without turning a narrow comparison into a false headline.

Where did the money come from — and why do reports say $10 million too?

You will see different amounts attached to this case. They describe different stages or measures; they should not be treated as interchangeable.

• More than $10 million was the amount alleged when prosecutors announced the indictment in September 2024. An initial charge is an allegation, not a final court finding.
• More than $8 million is the fraudulent royalty figure described by the DOJ after Smith's guilty plea and sentencing.
• $8,091,843.64 is the agreed forfeiture amount imposed by the judge. Forfeiture means surrendering assets or money connected with criminal conduct; it is not the same thing as a conventional fine or a verified personal net profit.
• Approximately $60,000 reportedly came from Spotify itself. Music Business Worldwide reported Spotify's statement that its prevention measures had limited Smith's royalties on that platform.

So no, Spotify did not pay him the entire $8 million. The case covers multiple platforms and royalty-payment arrangements. And saying Smith personally kept $8 million after expenses would go beyond what the judgment establishes.

Why fake streams take money away from legitimate artists

Music streaming is not generally a fixed-price transaction where every play earns the same number of cents. Services calculate different royalty pools under different plans, markets and rights arrangements. Money is then distributed according to eligible listening shares and applicable agreements.

The important part is the denominator.

Imagine a pool in which legitimate music has 900,000 eligible plays. Add 100,000 fraudulent plays that are treated as real. The legitimate catalog now accounts for 90% rather than 100% of the counted plays. That is a simplified illustration, not a calculation of the actual loss in Smith's case.

A fraudulent play can therefore transfer a fraction of a finite pool away from work people actually chose to hear. It also corrupts the data that services, distributors and artists use to make decisions. Spotify's royalty guide explains its recorded-music royalty model; the US Copyright Office's streaming-fraud inquiry specifically identifies royalty-pool dilution as a concern.

The defense argued that the damage to any individual artist was imperceptibly small. But distributing harm among many people does not make the combined amount disappear.

Why was the prison sentence only 18 months?

The parties proposed very different outcomes:

These comparisons were reported by Music Business Worldwide on October 7, drawing on sentencing submissions and coverage of the hearing.

The defense emphasized the lack of a measurable loss to any particular artist and requested a noncustodial sentence. Prosecutors wanted a stronger deterrent. The judge imposed a shorter prison term than either the guideline range or the probation recommendation.

That outcome does not establish that the financial damage was minor. Nor can a news report tell us every consideration behind the judge's decision without a complete statement of reasons.

Smith had originally faced three counts in the 2024 indictment. His 2026 guilty plea was to one count of conspiracy to commit wire fraud, carrying a statutory maximum of five years. Avoid describing the original counts as if he had been convicted of all three.

Is AI-generated music itself illegal?

No. Producing music with AI is not, by itself, the crime in this case.

The problem was the combination of false identities, automated plays, deceptive representations to platforms and fraudulent royalty collection. Copyright ownership, training data, licenses and voice imitation raise separate legal questions, but they were not the basis of this conviction.

This is also why blaming a specific music generator would be misleading. The indictment refers to an unnamed AI-music-company executive and collaborators; it does not establish that Smith used Suno or Udio.

I have written separately about AI music, creativity and human authorship and consent, credits and rights in AI music. Those questions matter. But they should not be confused with streaming bots impersonating listeners.

What are streaming services doing about it?

There is no single fix, but two approaches show where platforms are heading.

Spotify changed eligibility for recorded-music royalties in April 2024. A track must reach at least 1,000 streams within the previous 12 months to enter its recorded-music royalty pool calculation, alongside other eligibility conditions. Publishing royalties are calculated separately. The company explicitly said the change also targets a way of gaming the system through huge catalogs of barely played tracks.

That makes some mass-upload strategies less attractive. It does not mean any track that reaches 1,000 plays is genuine, or that fraud detection is solved.

Deezer reported a different pressure in July 2026: almost 90,000 fully AI-generated tracks arriving per day, representing more than half of new uploads at peak times in June. Yet AI-generated content accounted for just 1–3% of total listening, according to Deezer's own measurements. The company said it had classified up to 85% of streams on fully AI-generated tracks in 2025 as fraudulent and demonetized those streams.

Those figures are Deezer-specific platform statistics, not proof that 85% of all AI music everywhere is fraudulent. Deezer says it labels detected AI-generated material, excludes it from editorial and algorithmic recommendations, and removes material used for streaming fraud.

The technical challenge is not simply recognizing synthetic audio. A human can commit stream fraud with ordinary music, and an AI track can have genuine listeners. Services need to assess authenticity of activity, not just the origin of the recording.

The timing matters: Washington is now investigating streaming fraud

On October 7, 2026 — one day after Smith was sentenced — the US Copyright Office opened a public inquiry into music streaming fraud. It asks about bot farms, click farms, playlist stuffing, account hijacking, the effectiveness of countermeasures and possible industry or regulatory responses.

Initial public comments are due November 23, 2026; replies are due December 21, 2026.

An inquiry is not a new law. But it is a strong signal that the problem now extends beyond individual cases and into questions about how the industry should measure, identify and pay for listening.

And this was not the first streaming-fraud prison sentence anywhere. Danish prosecutors reported a separate appellate judgment in February 2025. The distinctive claim here is the first US criminal prosecution involving AI-assisted music streaming fraud, as described by the DOJ.

My takeaway: the exploit was the incentive system

The AI-generated tracks were inventory. The bot accounts were simulated demand. The royalty system converted that simulated demand into payments.

To me, that is the core of this story.

AI made it easier to supply enormous numbers of tracks, but the incentive to falsify engagement existed long before modern music generators. The operation depended on platforms accepting automated listening as genuine and allowing the resulting accounting data to move money.

The response should be better fraud detection, transparent accounting and meaningful consequences — without declaring every independent artist or AI-assisted creator suspicious by default.

Music can be synthetic. The audience cannot be fake if you expect to be paid for real listening.

Sources and fact-checking

• US Department of Justice — sentencing announcement, October 6, 2026 — sentence, scale, Taylor Swift comparison and forfeiture.
• US Department of Justice — guilty plea, March 19, 2026 — offense of conviction and agreed amount.
• United States v. Michael Smith — original indictment, 2024 (PDF) — account model, cloud setup and original allegations.
• Music Business Worldwide — sentencing report, October 7, 2026 — competing sentencing recommendations.
• Spotify for Artists — Royalties Guide — the 1,000-stream eligibility rule and its limitations.
• Deezer Newsroom — AI music uploads, July 21, 2026 — platform-reported AI-music and fraud statistics.
• US Copyright Office — Music Streaming Fraud inquiry, October 7, 2026 — official consultation and deadlines.
• Danish Prosecution Service — streaming fraud appellate judgment, February 5, 2025 — earlier case outside the United States.

Fact-checked October 14, 2026. Historical calculations and operational details attributed to the 2024 indictment are identified as allegations or Smith's own estimates where appropriate; they should not be mistaken for independent measurements.