Helen Smith, IMPALA’s executive chair, used the end of International Fraud Awareness Week to ask for tougher artist verification, know-your-customer checks and penalties for repeat offenders, in an op-ed on CMU today.
She has the remedy right, but the numbers show a crime that has changed shape: streaming fraud is now a volume game, and AI makes volume nearly free.
The clearest proof is an email from Michael Smith, the North Carolina man sentenced last month in the first US criminal prosecution of its kind. In 2018 he wrote that to avoid attention “we need a TON of content with small amounts of Streams,” as quoted by prosecutors.
| Measure | Figure | Source |
|---|---|---|
| Michael Smith’s early bot model | 1,040 accounts, about 661,440 streams a day | Prosecutors, via Music Business Worldwide |
| Implied plays per song per day | Roughly 2, across 300,000 songs | Music Business Worldwide |
| Spotify royalty threshold since April 2024 | 1,000 streams in the previous 12 months | Music Business Worldwide |
| Michael Smith’s sentence and forfeiture | 18 months and $8,091,843.64; prosecutors asked for at least 46 months | SDNY, via Music Business Worldwide |
| Fully AI tracks reaching Deezer, June 2026 | About 90,000 a day, over 50% of uploads at peak | Deezer |
| Same measure across 2025 | 13.4 million tagged, about 36,700 a day | Deezer, daily average calculated |
| Fraudulent share of streams on fully AI tracks, 2025 | Up to 85%, against 8% across the whole catalog | Deezer |
The model is simple: hundreds of thousands of tracks, each small enough to look innocent, which is exactly what generative tools now make for almost nothing.
Deezer says fully AI tracks make up only 1 to 3 percent of its streams, yet it calls their primary purpose fraud.
Thresholds stop the laziest version of this. By Music Business Worldwide’s arithmetic, Michael Smith’s early model averaged about 805 streams per track a year, below Spotify’s line, but prosecutors say he later ran as many as 10,000 accounts, which clears it.
Meanwhile every genuine small act under the threshold earns nothing, which is why IMPALA’s plan asks to reject payment thresholds for verified artists.
That is the independent’s problem in one line. IMPALA says independents have fewer resources to monitor fraud and can be “disproportionately affected,” and a filter built to catch spam will snag small honest catalogs first.
Deezer reports that its detector wrongly flags fewer than one in 10,000 human tracks, which is good, but a system that is 99.8 percent accurate still needs a working appeals desk.
IMPALA asks that no label playing fair lose repertoire or royalties “without prompt recourse options,” and that is the clause worth fighting for.
AI spam is not a creative movement that happens to attract fraud. It is the fraud. A catalog that grows by 90,000 tracks a day is not growing, it is being flooded.
Labels will not save listeners here, because 97 percent of people in Deezer’s Ipsos survey could not tell AI from human music in a blind test. The check has to happen at upload, and distributors, as IMPALA says, must verify their clients before delivery.
Enforcement is the weakest link. The first prosecution of its kind drew 18 months against at least 46 sought, and the defendant’s lawyers cited a September 2025 Spotify line that “all music is treated equally, regardless of the tools used.”
Treat that neutrality as a luxury. Until 90,000 daily AI uploads have to prove they are human, independents will keep paying for the defense and the fraudsters will keep paying nothing.




