A federal judge has sentenced North Carolina musician Michael Smith to 18 months in prison for an industrial-scale streaming fraud that paired hundreds of thousands of AI-generated tracks with as many as 10,000 bot accounts. The scheme produced billions of fake plays and more than $8 million in royalties, according to the U.S. Attorney’s Office for the Southern District of New York.
The October 6 sentence is the first in a U.S. criminal case centered on AI-assisted music-streaming fraud. It arrived just as the U.S. Copyright Office opened a formal inquiry into fake streams, bot farms, AI-generated music and possible responses from platforms, distributors, Congress and law enforcement.
Smith, 54, pleaded guilty in March to one count of conspiracy to commit wire fraud. U.S. District Judge John G. Koeltl also ordered him to forfeit $8,091,843.64 and serve two years of supervised release after prison. The 18-month term was below the 46 months prosecutors requested and the 24 months recommended by the U.S. Probation Office, according to Music Business Worldwide.
How the AI music streaming fraud worked
The operation ran from 2017 through 2024 and followed a three-part loop. Smith created thousands of accounts on Spotify, Apple Music, Amazon Music and YouTube Music. Software made those accounts continuously play tracks he controlled. The resulting royalties then flowed back to him.
The hard part was keeping that activity from looking like one impossibly popular song. Smith spread the automated plays across a huge catalog so each individual track generated less conspicuous traffic. Prosecutors said he used bulk email accounts, fictitious identities and fraudulently obtained debit cards to operate the listener accounts.
AI made the catalog cheap and fast to expand. Smith began working with the head of an unnamed AI music company and a music promoter in 2018, according to the original indictment summary. The supplier eventually delivered thousands of tracks a week. Smith assigned randomized song and artist names, then distributed the synthetic catalog across the streaming services.
The scale is easiest to see in one comparison from the case. In April 2023, Taylor Swift’s entire catalog received 9.3 million YouTube Music streams through family plans. Smith’s bot accounts generated 80.9 million family-plan streams for his AI catalog during the same month. That comparison covers a particular YouTube Music account type, not Swift’s total listening across every plan and platform.
The crime was the fraud, not making music with AI
The conviction does not establish that creating, uploading or earning money from AI-generated music is inherently illegal. Smith pleaded guilty to a conventional fraud charge because he fabricated listeners, misrepresented the activity to platforms and distributors, and collected money tied to plays that did not come from real consumers.
That distinction matters as services develop different approaches to synthetic music. Spotify, for example, has introduced labels and recommendation limits for AI-generated artist identities. Deezer tags fully synthetic tracks, excludes them from recommendations and demonetizes streams it determines are fraudulent. An artist’s use of an AI instrument, voice or production tool is a separate question from running accounts that manufacture plays.
AI’s role in Smith’s scheme was operational. Generative tools supplied the enormous number of distinct files needed to dilute the bot traffic across a catalog. In an email quoted by prosecutors, Smith wrote that he needed “a TON of songs fast” to work around platform anti-fraud systems. The audio was inventory for a traffic-manipulation business.
Why fake plays take money from real artists
Streaming services generally allocate a share of revenue to royalty pools, then distribute those pools according to listening activity and contractual rules. A fake play can therefore do more than trigger an undeserved payment. It can marginally reduce the share available for legitimate listening, distort charts and recommendations, and contaminate data used for marketing, touring and artist development.
The Copyright Office’s October 7 notice describes a broader set of tactics than Smith used: bot and click farms, playlist stuffing, account hijacking, misleading marketing services, manipulated copies of existing recordings and “ghost tracks” that falsely claim another work. It also notes that services examine signals including listening time, geography, an account’s concentration on particular artists, and the relationship between streams, saves and follows.
Generative music raises the volume of material that those systems must assess. Deezer reported in April that it was receiving nearly 75,000 fully AI-generated tracks a day, about 44% of its daily uploads. Those tracks accounted for only 1% to 3% of listening, but the company classified 85% of their streams as fraudulent in 2025 and excluded those plays from royalty calculations. Those are Deezer’s own figures and should not be treated as an industry-wide measurement, but they show why catalog scale and listening authenticity have become linked problems.
What the Copyright Office is asking
The new inquiry asks for evidence on the prevalence and economic impact of streaming fraud, including its relationship to AI-generated music. It is also examining whether industry practices inadvertently create incentives for manipulation and whether alternative royalty models could reduce them.
Several of the questions are practical rather than abstract. The office wants information on trusted-distributor programs, content and customer verification, royalty clawbacks, fines, shared suspicious-activity reports, standardized metadata and a possible database of known abusers. It also asks whether content that is ineligible for statutory royalties should be labeled and how criminal enforcement could better address organized operations.
That focus matches the industry’s most recent response. In September, IFPI and a group of labels and distributors launched a Streaming Integrity Initiative built around identity and rights verification, content vetting, AI-related risk checks, action against repeat offenders and legally permitted intelligence sharing.
No single control closes the loop. Audio detection may flag synthetic content without proving that its plays are fake. Traffic analysis can identify coordinated listening without resolving who owns the recording. Stronger identity checks can deter repeat offenders but may add friction for legitimate independent artists. The Smith case shows why platforms and distributors need signals from all three layers: who uploaded a track, what the audio and metadata contain, and how accounts behave after release.
What happens next
The Copyright Office will accept initial comments through November 23 and replies through December 21. The proceeding does not itself change streaming law or platform rules. It will build a record for Congress on whether voluntary coordination, civil remedies, information-sharing rules or new legislation are warranted.
Smith’s sentence provides a clear enforcement precedent, but it also documents an old weakness in a newer form. Bot-driven plays existed before generative music. AI did not invent the fraud; it removed the catalog-production bottleneck that once made this kind of operation harder to scale and easier to spot.
