AI Music Streaming Fraud: What the $8M Bot Case Actually Punished

On 6 October 2026, in Manhattan federal court, Judge John G. Koeltl sentenced Michael Smith, 54, of Cornelius, North Carolina, to 18 months in prison for a scheme that used generative AI and an army of bot accounts to extract streaming royalties for seven years. It is the first case of its kind to reach sentencing in the United States, and the Justice Department's own framing calls it the first criminally charged "super intelligence-assisted" music streaming fraud.

AI music streaming fraud is now a sentenced federal offence, and that makes the exact wording of the charge worth more than any headline about it. Smith pleaded guilty to one count of conspiracy to commit wire fraud. He did not plead guilty to generating music with AI, to uploading it, or to selling it — because none of those is a crime. The distinction is not pedantry. If you release AI-assisted music, the difference between what was punished here and what you do every week is the whole question, and almost no coverage of the sentence draws it.

What the court actually decided

The sentence is lighter than the government wanted and heavier than the defence asked for. Here is the record in one place, as reported consistently by Music Business Worldwide, Music Ally, Rolling Stone and Digital Music News, and announced by the US Attorney's Office for the Southern District of New York.

The sentencing positions are the part that shows how the court weighed it. Prosecutors sought at least 46 months, the bottom of a 46-to-57-month guideline range. The US Probation Office recommended 24 months. Smith's lawyers asked for probation with no custody at all. The judge landed at 18 months — below the probation recommendation and well below the government's request, on a count carrying a five-year maximum.

US Attorney Jamie McDonald's office described the conduct in terms that are worth reading closely: Smith "exploited super intelligence technology to generate a fraud," and did so "by flooding music streaming platforms with automated bots in the place of consumers." The object of the sentence is the bots. The technology is the instrument.

He was not convicted for making AI music

Every outlet headlined this as AI music fraud, and the shorthand has consequences: it leaves anyone making music with Suno or Udio with the impression that the activity itself has been criminalised somewhere. It has not. What the case establishes is narrower and more useful.

The elements of wire fraud are a scheme to obtain money by deception using interstate wires. The deception here was representing automated playback as listening by real consumers. Swap the AI catalogue for a catalogue of silence, field recordings or thirty-second loops and the offence is identical — which is, in fact, what most pre-AI streaming fraud used.

The AI mattered for one reason: supply. The indictment describes hundreds of thousands of AI-generated songs, sourced with the help of the chief executive of an AI music company, who supplied tracks at volume. Generative tools solved the catalogue problem, and the catalogue problem was the binding constraint on the fraud. That is the honest description of AI's role, and it is a long way from "AI music is illegal now."

Three money figures, three meanings

If the number you have seen for this case is $10 million, $14 million or $8 million, all three are defensible and they are not the same quantity. Coverage substitutes one for another freely.

The spread is not sloppiness. A charging document alleges; a sentencing letter totals what arrived; a forfeiture order reflects what the parties could agree was loss. The $8.09 million figure is the one with legal force, and it is the smallest of the three — which is why "the $8 million fraud" and "the $14 million fraud" describe the same man in the same week.

How the bots were meant to stay invisible

The intuition most people have about streaming fraud is that it drives one track to enormous numbers. Smith's plan was the exact inverse, and his own words are the clearest statement of the strategy anywhere in the record. In an October 2018 email he wrote that "in order to not raise any issues with the powers that be we need a TON of content with small amounts of Streams."

That sentence is the design document. Detection at the time looked for anomalies on individual tracks — a song with implausible velocity, a release with a sudden unexplained spike. A catalogue of hundreds of thousands of songs, each accumulating a handful of plays a day, produces no such signal anywhere while producing an enormous total.

The arithmetic of the early configuration, as laid out by Music Business Worldwide from the government's filings:

Per-track figures depend on how large the catalogue was at any moment, which the filings do not fix. MBW's working assumption of roughly 300,000 songs yields about two plays per song per day — arithmetic on an assumed denominator, not a prosecution finding, so read it as illustrative. The direction is what counts: deliberately, defensibly small per song.

The indictment's own phrasing confirms the scale ran in parallel rather than in sequence — Smith "used over a thousand bot accounts simultaneously to artificially boost streams of his music across the Streaming Platforms." Four platforms, a thousand or more accounts at once, for seven years.

The 1,000-stream rule that now works against it

Here is where the case stops being a crime story and starts being relevant to your release. Spotify's track monetisation eligibility documentation sets out a rule announced in November 2023 and in force for all artists from April 2024:

"Starting in April 2024, tracks must have reached a threshold of at least 1,000 streams in the previous 12 months to be included in the recorded music royalty pool calculation."

There is also a minimum number of unique listeners, which Spotify declines to publish — explicitly, in its own words, "to prevent further manipulation by bad actors" — on the basis that a published figure could be gamed by streaming a track repeatedly to qualify.

Now read that against the thin-spread strategy. A track accumulating two plays a day accrues roughly 730 a year, which does not clear 1,000. The threshold is, structurally, the answer to exactly the playbook Smith ran: it makes a catalogue of barely-streamed tracks worth nothing, no matter how many of them there are, and it forces any thin-spread scheme to scale its bot fleet until each track crosses the line — at which point the fleet itself becomes the detectable thing.

The honest limit, which the timeline forces: this rule did not catch him. It took effect in April 2024, at the very end of a scheme that ran from 2017, and it was not the mechanism that unravelled the case. Nor was it introduced in response to this prosecution. It is the structural answer arriving after the fact — and it lands on everyone. The same rule that strands a fraudulent catalogue strands an honest small one, which is why a release with 300 genuine plays earns zero recording royalty on Spotify. Our streaming royalties breakdown has the per-platform rates, and the 1,000-stream maths is worked through there.

What the platforms changed after the flood

The enforcement context around this case is public, dated and quotable, and it explains why a legitimate AI release now meets friction that did not exist three years ago. Spotify's own policy post of 25 September 2025, Spotify strengthens AI protections, gives both the scale and the motive:

"In the past 12 months alone, a period marked by the explosion of generative AI tools, we've removed over 75 million spammy tracks from Spotify."

Note the date on that figure — it covers the 12 months to September 2025, which makes it a year old now, and it is frequently quoted without one. The same post states the economic reason bluntly: total music payouts grew from $1 billion in 2014 to $10 billion in 2024, and "big payouts entice bad actors."

Deezer's numbers show why screening exists at all: fully AI-generated tracks now approach half or more of everything delivered to it, and it reports around 85% of the streams those tracks receive as fraudulent, which it demonetises. Our Deezer AI music policy page covers how that tagging works in practice.

One thing the post does not do is conflate transparency with enforcement. Spotify says the disclosure work "is not about punishing artists who use AI responsibly or down-ranking tracks for disclosing information about how they were made," and concedes the limit of any such system: because it depends on the artist, "the absence of a credit doesn't mean AI wasn't used."

Fraud screening and AI screening are not the same test

This is the practical confusion the headlines create, and it is worth separating cleanly. Two different systems look at your release, with different inputs, different triggers and different consequences.

Neither system cares what the other one found. A fully human recording gets its royalties withheld if someone buys it bot streams; a fully AI-generated track with honest listening passes fraud screening and may still be flagged, labelled or rejected on the audio. How the detection side works — and what it actually measures — is covered in our AI music detector explainer.

The live risk for a legitimate AI releaser is not criminal exposure. It is buying promotion that turns out to use bots, because platforms treat bought streams as artificial whatever you believed you were buying. Deezer, Spotify and Tidal all withhold royalties on streams identified as artificial, and good intentions are not a defence in any of those policies. A service promising guaranteed play counts is the shape of the problem.

Worth saying plainly, because the coverage invites the opposite inference: distributors do not ban AI music. Six accept it openly — DistroKid, RouteNote, UnitedMasters, LANDR, Amuse and Symphonic. What happens to AI releases is that some get flagged and rejected by automated screening, which is a different and much more specific problem. Our DistroKid AI music policy and Spotify AI music policy pages go platform by platform, and eraseai.co's guide to AI music removed from streaming covers the removal causes in more depth.

What it actually costs you

The royalty pool is finite and shared. Spotify's stated rationale for the spam filter is that these behaviours "dilute the royalty pool and impact attention for artists playing by the rules" — so a scheme like this draws from a pot you draw from too. That is the direct cost, and over seven years it is real.

The indirect cost is harder to undo: the 1,000-stream rule, the spam filter, the upload screening, the payouts withheld pending review. None of it was built with you in mind, and all of it is now in your path.

But it would be dishonest to present fraud enforcement as the thing standing between you and a career, because it is not. The binding constraint is listeners. Luminate counted 253 million tracks available in 2025, of which 88% got fewer than 1,000 plays in the year. Clearing a distributor's screening, clearing the monetisation threshold and clearing fraud detection still leaves you with the actual problem, which is that nobody knows the track exists. We put the numbers on that in nobody listens to your AI music, and it remains the least comfortable page on this site.

Which reframes what to do once a track is finished. Algorithmic discovery is not a plan you control; the routes that do not depend on it are getting placed and selling direct. Paid sync — television, film, games, advertising — is where the money conversation in this niche actually is, and played.fm exists for that route and for selling direct to the listeners you do reach. Neither solves discovery. Both work without it, which is a more honest proposition than a growth channel. Erasy's AI music earnings breakdown puts numbers on a reported $15,000 month.

The case ends with one man going to prison for eighteen months over $8.09 million, and an industry that rebuilt its plumbing around the possibility of him. The IFPI called the result a clear message that streaming fraud is a crime with serious consequences, while noting enforcement is only one part of the answer. For everyone releasing honestly, the enforcement was never the part that affected you. The plumbing is.