How much did Michael Smith make from AI streaming fraud?

Prosecutors say he collected more than $8 million in royalties, and he agreed to forfeit $8,091,843.64. Earlier reporting on the 2024 indictment described the scheme as worth more than $10 million.

What sentence did Michael Smith get?

On 6 October 2026, US District Judge John G. Koeltl sentenced him to 18 months in prison and two years of supervised release. He had pleaded guilty in March 2026 to conspiracy to commit wire fraud.

How do streaming services detect bot streams?

Platforms look for coordinated behavior across accounts, such as shared payment methods, IP and device overlap, round-the-clock sessions and listeners who stream almost entirely from one catalog. Spreading plays across many tracks defeats per-song spike rules, so detection has to happen at the account-cluster level.

AI Streaming Royalty Fraud: Michael Smith Pleads Guilty

On 19 March 2026, North Carolina musician Michael Smith pleaded guilty in federal court in Manhattan to a scheme that used hundreds of thousands of AI-generated songs and thousands of bot accounts to collect more than $8 million in streaming royalties. Prosecutors describe it as the first criminal case of its kind in the US. For developers it is a clear example of how cheap generative content and basic automation can defeat platforms' payout and fraud-detection logic.

What happened

Smith, 54, of Cornelius, North Carolina, pleaded guilty before US District Judge John G. Koeltl to one count of conspiracy to commit wire fraud, which carries a maximum of five years. When he was indicted in September 2024 he faced three felony counts: wire fraud, wire fraud conspiracy and money laundering conspiracy. Prosecutors had originally put the scheme at more than $10 million. The plea agreement sets forfeiture at $8,091,843.64, which is why this headline uses $8 million.

According to the US Attorney's Office for the Southern District of New York, the scheme ran from 2017 to 2024:

Smith obtained hundreds of thousands of AI-generated tracks. Earlier reporting said he worked with the CEO of an AI music company and a music promoter, who are unnamed co-conspirators.
He created thousands of accounts on Amazon Music, Apple Music, Spotify and YouTube Music, at times running up to 10,000 at once. He registered them with fake email addresses and fraudulently obtained debit cards.
Software streamed his catalog from those accounts continuously. He spread plays across thousands of tracks so that no single song showed a suspicious spike.

The bot accounts streamed the songs billions of times. In one month of 2023 they generated 80.9 million streams. Update: on 6 October 2026, Judge Koeltl sentenced Smith to 18 months in prison followed by two years of supervised release.

Who is affected

Artists and rights holders. Streaming royalties are paid from a shared pool divided by share of streams, so fraudulent plays take money from legitimate artists.
Streaming platforms and royalty collectors, including the Mechanical Licensing Collective, which commented on the plea.
Anyone building pay-per-engagement systems: ad networks, creator funds, affiliate programs and API marketplaces that pay out on usage.

What it means for developers and engineering teams

This case is about economics more than AI. Streaming payouts rest on one assumption: a stream represents a person listening. Smith broke it in two ways. Generative tools made it practically free to create a huge catalog, and automation made it cheap to fake listeners. The important design choice was spreading plays thinly. Fraud rules that flag one track with abnormal plays, or one account with abnormal hours, don't trigger when every track gets a modest, plausible number of streams.

The detection signals that hold up against that strategy work across the graph, not per item:

Account clustering: shared payment instruments such as prepaid debit cards, email naming patterns, signup timing, device fingerprints, and IP or ASN overlap (VPN and datacenter ranges).
Listener-to-catalog concentration: a set of accounts that streams almost entirely from one rights holder's catalog is suspicious even if each track's numbers look normal.
Behavioral uniformity: 24/7 sessions, no skips, no searches, no playlist edits.
Catalog signals: huge upload volumes from one distributor account with near-zero organic discovery.

Any system that pays for engagement should assume the engagement can be manufactured. Build graph-level anomaly detection and hold payouts in escrow long enough to run it.

What to do now

If you run a payout system, model fraud as coordinated rings. Compute features per cluster of accounts (shared cards, IP ranges, signup cohorts), not only per account or per item.
Delay and claw back: hold payouts for a review window and keep contractual rights to reverse fraudulent earnings.
Add friction where it costs bots more than humans: payment-instrument verification, device attestation, and rate limits on concurrent sessions per payment method.
Track catalog velocity: alert when an uploader's catalog grows by orders of magnitude while organic discovery stays flat.
If you publish AI-generated content, read each platform's terms. Several now restrict or label AI-generated uploads, and the legal exposure here came from faking the listeners, not from making the songs.

Musician pleads guilty to $8M AI-generated music streaming royalty fraud

What happened

Who is affected

What it means for developers and engineering teams

What to do now

Sources

Get the next deep dive in your inbox

See what launched this week

The weekly digest