The Volokh Conspiracy
Mostly law professors | Sometimes contrarian | Often libertarian | Always independent
Prosecutors Allege $10M Royalty Scam Using AI-Generated Songs
The N.Y. Times (Maia Coleman) reports:
A North Carolina man used artificial intelligence to create hundreds of thousands of fake songs by fake bands, then put them on streaming services where they were enjoyed by an audience of fake listeners, prosecutors said.
From the Indictment:
From approximately 2017, up to and including 2024, MICHAEL SMITH, the defendant, orchestrated a scheme to steal millions of dollars of musical royalties by fraudulently inflating music streams on digital streaming platforms (the ''Streaming Platforms"), such as Amazon Music, Apple Music, Spotify, and YouTube Music. SMITH purchased from a coconspirator hundreds of thousands of songs that were created through artificial intelligence ("AI") and then uploaded to the Streaming Platforms.
SMITH then used "bots"—automated programs—to stream the AI-generated songs billions of times. At the height of his fraudulent scheme. SMITH used over a thousand bot accounts simultaneously to artificially boost streams of his music across the Streaming Platforms. By manipulating the streaming data in this manner, SMITH fraudulently obtained more than $10 million in royalty payments to which he was not entitled….
As alleged herein, MICHAEL SMITH, the defendant, made false and misleading statements to the Streaming Platforms, Rights Organizations, and certain companies that facilitate the distribution of artists' music to the Streaming Platforms. As described below, those lies were repeated and varying but all were intended to promote and conceal his massive streaming manipulation fraud. As a result of his false and misleading statements, SMITH fraudulently obtained millions of dollars in royalty payments from the Streaming Platforms, Rights Organizations, and music distribution companies. Those funds ultimately should have been paid to the Songwriters and Artists whose works were streamed legitimately by real consumers….
At certain points. SMITH had as many as 10,000 active Bot Accounts on the Streaming Platforms. Signing up such a voluminous number of Bot Accounts on the Streaming Platforms was labor-intensive, and SMITH paid individuals located abroad as well as coconspirators located in the United States to do the data entry work of signing up for the Bot Accounts. For example, in a May 11, 2017 email to a coconspirator ("CC-1"), SMITH asked CC-1 to create Bot Accounts on a particular Streaming Platform: "Make up names and addresses[.] [J]ust make sure they all are the same for family member and also make sure everyone is over 18." …
Early in the scheme, SMITH used the catalog of a music publicist ("CC-2") to fraudulently generate royalty payments. Later, SMIIB attempted to sell his fraudulent streaming scheme as a service, in which other musicians would pay him for streams he would fraudulently generate or share royalties with him in exchange for fraudulent streams of their music. But neither strategy allowed SMITH to gain access to the massive volume of songs the scheme needed in order to evade detection and succeed on a large scale….
[Smith] eventually turned to artificial intelligence to expand his fraudulent scheme, and in tum, his illicit proceeds. In or about 2018, SMITH began working with the Chief Executive Officer of an Al music company ("CC-3") and a music promoter ("CG-4") to create hundreds of thousands of songs using artificial intelligence that SMITH could then fraudulently stream….
CC-3 ultimately provided MICHAEL SMITH, the defendant, with hundreds of
thousands of Al songs for which he could manipulate the streams. CC-J's songs were typically given file names that were a randomized list of letters and numbers, such as "n_7a2b2d74-1621-4385-895d-bl e4af'78d860.mp3." SMITH then created randomly generated song and artist names for audio files so that they would appear to have been created by real artists rather than artificial
intelligence. For example:a. An alphabetically consecutive selection of25 of the names of the AI songs SMITH used is as follows: "Zygophyceae," "Zygophyllaceae," "Zygophyllum." "Zygopteraceae," "Zygopteris," "Zygopteron," "Zygopterous," "Zygosporic," "Zygotenes," "Zygotes," "Zygotic," "Zygotic Lanie," "Zygotic Washstands," "Zyme Bedewing," "Zymes," "Zymite," "Zymo Phyte," "Zymogenes," "Zymogenic," "Zymologies," "Zymoplastic," "Zymopure," "Zymotechnical,"
"Zymotechny," and "Zyzomys."b. An alphabetically consecutive selection of 25 of the names of the "artists" of the AI songs SMITH used is as follows: "Calliope Bloom," "Calliope Erratum," "Callous," "Callous Humane," "Callous Post," "Callousness," '·Calm Baseball," "Calm Connected," "Calm Force," "Calm Identity," "Calm Innovation," ''Calm Knuckles," "Calm Market," "Calm The Super," "'Calm Weary," ''Calms Scorching," "Calorie Event," "Calorie Screams,'' "Calvin Mann," "Calvinistic Dust," "Calypso Xored," ''Camalus Disen," "Camaxtli Minerva," "Cambists Cagelings," and "Camel Edible." …
The Al technology that CC-3 used to generate AI songs for MICHAEL SMITH, the defendant, improved over time, making it less likely that the Streaming Platforms would detect the scheme….
MICHAEL SMITH, the defendant, made numerous misrepresentations to the Streaming Platforms in furtherance of the fraud scheme. For example, SMITH repeatedly lied to the Streaming Platforms when he used false information to create the Bot Accounts and when he agreed to abide by terms and conditions that prohibited streaming manipulation. SMITH also deceived the Streaming Platforms by making it appear as if legitimate users were in control of the Bot Accounts and streaming music when, in fact, the Bot Accounts were hard-coded to stream SMITH's music billions of times. SMITH also caused the Streaming Platforms to falsely report billions of streams of his music, even though SMITH knew that those streams were in fact caused by the Bot Accounts rather than real human listeners….
I could totally imagine those as band names.
Editor's Note: We invite comments and request that they be civil and on-topic. We do not moderate or assume any responsibility for comments, which are owned by the readers who post them. Comments do not represent the views of Reason.com or Reason Foundation. We reserve the right to delete any comment for any reason at any time. Comments may only be edited within 5 minutes of posting. Report abuses.
Please
to post comments
Gotta admire the crook's ingenuity.
Heh. I had two tabs open, and confused which was which, and so my thought upon reading your comment was, "What did Trump do this time?" Glad to see this is a nice, apolitical crime.
Well, stealing money is a motivation as old as time.
Crook?
Seems to me he was just following the rules as he found them.
I hope you aren't going to try to tell me bots can't identify as humans.
"Michael Smith" sounds suspiciously ... AI generated. Is this how AI starts self-funding electrical bills?
Nah. Selling drugs is much more lucrative.
I read the indictment with interest, but was shocked by the prosecutor’s innumeracy, manifested in the sentence, “A billion fake streams spread across tens of thousands of songs, however, would be more difficult to detect, because each song would only be streamed a couple of times” (p. 8, near the bottom).
In the introduction to “A Distant Mirror”, her history of the 14th century, Barbara Tuchman wrote, “[T]he chroniclers did not use numbers as data but as a device of literary art to amaze or appall the reader.” I get the impression that our prosecutor hadn’t got much farther in his mathematical education…
Completely off topic, but I really enjoyed "A Distant Mirror." Although it was dense enough that about halfway through, I moved from print to audio to finish it up.
"I could totally imagine those as band names."
Especially Callous Post and Camel Edible. In fact, if VC bloggers ever form a band, I would suggest Callous Post as a name.
"Calvinistic Dust" - once cleaned, always cleaned
More like, fated to be cleaned.
Cleaned; whether you want it or not.
I don’t think this is limited to fake videos. Some time back, when a billion views was the world record, there was a new JLo video with 500 million views. I watched, but it sucked, no hook, no nothing. Usually videos by big singers that suck get a few million or less. You can’t get 500 million without multiple repeated views, which I was certain would not happen.
Maybe I had a tin ear for that particular song. Maybe not.
Was she shaking anything in the video?
"Later, SMIIB attempted to sell his fraudulent streaming scheme as a service"
What happened here? Smiib?
That music's not going to listen to itself. Oh wait...
This for some reason reminded me of a Frederick Pohl short story called "The Midas Plague"
"In a world of cheap energy, robots are overproducing the commodities enjoyed by humankind. The lower-class "poor" must spend their lives in frantic consumption, trying to keep up with the robots' extravagant production, while the upper-class "rich" can live lives of simplicity. Property crime is nonexistent, and the government Ration Board enforces the use of ration stamps to ensure that everyone consumes their quotas. The story deals with Morey Fry, who marries a woman from a higher-class family. Raised in a home with only five rooms she is unused to a life of forced consumption in their mansion of 26 rooms, nine automobiles, and five robots, causing arguments. Trained as an engineer, Morey modifies his robots to enjoy helping to consume his family's quota. He fears punishment when his idea is discovered, but the Ration Board—which has been looking for a way to abolish itself—quickly implements Morey's idea across the world."
Here's a link if anyone wants to read it:
https://archive.org/details/galaxymagazine-1954-04/page/n7/mode/2up?view=theater
Seems more like a problem with the business model than how it was used.
This alleged scheme highlights what I’ve long seen as one of the flaws in the general model used for streamed music revenue sharing. That flaw is the pooling of revenues, e.g. from an entire streaming service, (a portion of) which are to be split among various rights-holders. I think the better model would be to split that portion of revenues on a per-customer basis.
An overly-simplified hypothetical to demonstrate what I mean:
A hundred people sign up for McIntosh Music and pay their $10 a month so that they can stream Bobby Streisand’s entire music catalog. They don’t stream a lot, only about 10 songs a day, but the service is still worth the money to them.
A hundred other people sign up for McIntosh Music and pay their $10 a month so that they can stream Tony Swift’s entire music catalog. They stream a lot, about 100 songs a day.
As I understand the current basic model, though I’m sure there are a number of winkles and negotiated incentives, the $20 a month is pooled and 60 or 70% of it goes to the rights-holders based on the portion of total streams the respective rights-holders’ music accounted for. (That revenue is also split between publishing rights-holders, who get a smaller portion, and recording right-holders who get a larger portion.)
For this hypothetical I’ll say Mr. Streisand and Mr. Swift both own all of the publishing and recording rights associated with their respective music catalogs.
So Mr. Streisand gets around $1.20 a month and Mr. Swift gets around $12 a month even though interest in the former’s music effectively generated the same amount of revenue as interest in the latter’s did. (There’s a lot more potential nuance I won’t get lost in.)
I’m sure there arguments to be made in favor of the general model as it exists, but I’d prefer the per-stream splitting be done on a per-customer (and even per-customer dollar) basis. If I pay $10 a month and split my streaming roughly evenly between Muriel Haggard, Michelle Jackson, and Dana Henley, then I’d like to think that each of them was getting around $2 a month from my subscription payment.
At any rate, this alleged scheme works because the revenue is pooled before being split based on streams. The bots streaming far more than average means that they effectively direct more revenue to their operators than those operators are contributing to the pool through subscription fees. This imbalance is increased by the use of family plans which allows even more streaming per subscription dollar spent.
Fundamentally, when the people who pay are different from the people who consume, a big opportunity for corruption exists.
This isn’t really conceptually different from insurance schemes, where doctors generate fake records of fake procedures given to fake patients in order to bill the government or insurance companies. Or welfare fraud, when people fake poverty or create fake poor people out of whole cloth. And before that there had long been people who staged fake auto accidents, committed arson, etc. in order to get an insurance payout.
While AI can be a fantastic tool for generating content, there’s always the risk of misuse, like we saw in this situation. It’s a clear reminder that we need to be cautious and fully understand how AI is being applied, especially in areas like music production and managing royalties. If you want to dive deeper into AI and its various applications, websites like https://ai-depot.net/ have some really in-depth articles on the subject. Getting a grip on how these technologies work is crucial, not just to avoid legal issues but to use them ethically and responsibly. As AI continues to grow, it’s important for artists and companies to stay informed about copyright laws to protect their own work and respect the rights of others.