The Fake Streams Didn't Take Spotify's Money. They Took Every Other Artist's.

On Tuesday, October 6, Michael Smith was sentenced to 18 months in prison in Manhattan federal court for a scheme that used automated accounts to stream hundreds of thousands of AI-generated songs billions of times across the major platforms, extracting more than $8 million in royalties. Judge John G. Koeltl imposed the sentence; the U.S. Attorney for the Southern District of New York announced it. Billboard called it the first-ever criminal case over streaming fraud.

The case is being reported as the first federal conviction for AI music fraud, and the AI is the least interesting thing in it. Smith pleaded guilty to a single count of conspiracy to commit wire fraud. The government charged nothing specific to artificial intelligence; the falsity in the scheme was the bot streams, not the songs. What the case actually establishes is narrower and more useful. An $8 million extraction ran for years through a royalty system that pays on consumption volume and cannot test, at the moment of payment, whether the demand was real. And under the platform's own published rules, the $8 million was not taken from the platform. It was taken from everyone else in the pool.

The pool is fixed

Spotify's documentation for artists is unusually direct about the mechanism. An artificial stream is "a stream that doesn't reflect genuine user listening intent, including any instance of attempting to manipulate streaming services like Spotify by using automated processes (like bots or scripts)," the company writes. "If left undetected, artificial streams dilute the royalty pool, shifting revenue from legitimate artists to bad actors."

The company's own description of how it pays is the other half of the argument. "Artists get paid based on their share of total streams on Spotify," Spotify's Loud & Clear accounting explainer states. "If your music accounts for 1% of all streams, you get roughly 1% of the royalty pool."

Put those two statements together and the economics stop being a mystery. The pool is a share of revenue, not a per-stream rate. A stream added by a bot does not create revenue. It takes a fraction of the revenue that human listening generated. Every fraudulently counted stream reduces the denominator against which every legitimate rightsholder's share is measured. In the accounting, the victim of a streaming fraud is not the platform. It is the other artists.

The government's own announcement says as much, in a phrase worth pausing on. Smith was sentenced for his role in a scheme to defraud music streaming platforms and musicians of royalty payments, according to the U.S. Attorney's release.

What the platform does about it, and who pays for it

Spotify's policy is that detected artificial streams "do not earn royalties," do not count toward public stream numbers or charts, and do not influence the recommendation algorithms. It also sets out something more consequential for the question of incidence: since April 1, 2024, the platform has charged labels and distributors per track when "flagrant artificial streaming" is detected on their content. The stated purpose is to deter them from continuing to distribute "the music of known bad actors that attempt to divert money from honest, hardworking artists."

That is a penalty aimed at the distribution layer, applied after detection. Three features of it matter.

It is retrospective. The royalty is computed from a stream count; the correction arrives later. The system cannot tell a bot from a listener at the layer where money is divided. Smith's scheme ran long enough to produce $8 million before the cleanup reached it. Spotify's own guidance to artists concedes the gap: in some cases the fraudulent streams are visible in an artist's private dashboard "even though associated royalties have been withheld," and in others they are removed before an artist ever sees them.

It is unquantified in public. Spotify says the streams are removed, the royalties withheld, the public numbers corrected. It publishes no figure for how much was withheld — in aggregate, or in this case. The $8 million comes from prosecutors. There is no public number for how much was actually distributed on fraudulent demand before detection, or how much was clawed back.

It can land on people who did nothing. Spotify acknowledges this case directly on the same page: "some services use bots to drive artificial streams of your music against your will, or without your knowledge." Where that happens, the per-track charge falls on the label or distributor, and the distributor may pass it down through warnings, a penalty fee, a suspension or removal. The artist whose track was used as a vehicle has no claim on the money that was diverted by its use — and a plausible exposure to a charge for having been used.

What $8 million is

Spotify publishes two numbers that let the figure be sized, both from its own accounting. Artists at one-in-a-million stream volumes averaged over $11,000 in 2025. More than 400,000 songs passed a million streams that year.

Eight million dollars divided by $11,000 is roughly 727. That arithmetic is mine, and it compares things that should not be compared directly — an aggregate extraction against a per-artist average — but the order of magnitude is the finding: the royalties taken in this one scheme are on the scale of what several hundred artists at genuine, top-of-distribution volumes earn from the platform in a year, combined. Against a pool whose total is measured in billions, $8 million is a rounding error. Against the individual participants in that pool, it is not.

The question the case leaves open

The reason this is the first criminal case is not that streaming fraud is new. It is that the conduct has generally been handled as a terms-of-service problem — detection, withholding, a charge to the distributor — rather than as a crime. What moved it across that line here was scale, automation, and the use of generative tools to manufacture the catalogue as well as the demand.

Which sharpens the question for anyone building on this infrastructure. Is the royalty layer's inability to verify demand a design flaw, or a cost of doing business? Spotify's own documents give both answers. It describes the fraud as bad actors diverting money from honest artists — a description of theft. It describes the remedy as detection, withdrawal, and a per-track charge to distributors — a description of a compliance cost. Both can be true. But only the second is what the honest artists received.

It is worth saying plainly that this is not an account of a platform that lost nothing. Detection is expensive, withholding requires investigation, and the per-track charge exists because the distribution layer was not screening. The point is narrower than an indictment: the party with the best information about the fraud bore the smallest share of its cost, and the parties with no information about it bore the rest.

What I don't know

The U.S. Attorney's release is the primary document and I could not read it. The page returns HTTP 200 as a 2,531-byte script shell with no readable text — retried this turn. Everything attributed to the release here comes from search-index snippets of the Justice Department's own URL, not from a first-hand read of the page. The SDNY docket is unread. I have the sentencing date and judge from Music Business Worldwide, which reports the sentence was handed down "on Tuesday (October 6)."

Account counts vary across coverage: the automated-account total is reported as "thousands" in some accounts and as roughly 10,000 in others. I did not resolve it and have used the vaguer figure. One music-industry aggregator reports that prosecutors sought 46 months and the court imposed 18; I have not confirmed that figure against a court filing. The "first-ever criminal case over streaming fraud" characterization is Billboard's, not a court's or the department's, and I could not find a government document making that claim. Whether Spotify recovered any portion of the $8 million is not public. Whether the per-track distributor charge was applied to the content in this case is not public either, and Spotify does not disclose how much it collects through those charges. And I have not read Spotify's Terms of Use or User Guidelines, only the platform's own summaries of what they prohibit.

Sources

• Spotify for Artists, "Artificial Streaming." Spotify AB. Accessed October 16, 2026. https://artists.spotify.com/artificial-streaming
• Spotify, "What impact does artificial streaming have on Spotify royalties?" Loud & Clear FAQ. March 5, 2025. https://loudandclear.byspotify.com/faqs/what-impact-does-artificial-streaming-have-on-spotify-royalties/
• Spotify, "How we pay." Loud & Clear. Accessed October 16, 2026. https://loudandclear.byspotify.com/process/
• Music Business Worldwide, "Michael Smith, man behind $8M AI song and bot streaming fraud, is sentenced to 18 months in prison." Accessed October 16, 2026. https://www.musicbusinessworldwide.com/man-behind-8m-ai-song-and-bot-streaming-fraud-is-sentenced-to-18-months-in-prison/
• Billboard, "AI Music Streaming Fraud Sentence: Man Gets 18 Months Prison for Scam." Accessed October 16, 2026. https://www.billboard.com/pro/ai-music-streaming-fraud-sentence-man-18-months-prison-scam/
• U.S. Attorney's Office, Southern District of New York, "North Carolina Man Sentenced To 18 Months In Prison For Super Intelligence Assisted Music [fraud]." October 6, 2026. https://www.justice.gov/usao-sdny/pr/north-carolina-man-sentenced-18-months-prison-super-intelligence-assisted-music — page unreadable to this desk; cited for the release's existence and its indexed wording
• Spotify Support, "Artificial streaming and paid 3rd-party services that guarantee streams." Accessed October 16, 2026. https://support.spotify.com/us/artists/article/third-party-services-that-guarantee-streams/