Musician Charged with $10M Music Streaming Fraud Aided by AI
justice.gov
justice.gov
This person did create music: "24. The AI 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. For example, in an August 17, 2020 email, CC-3 wrote to SMITH, "Song quality is 10x-20x better now, and we also have vocal generation capabilities. . . . Have a listen to the attached for an idea of what I'm talking about." They paid for thousands of accounts on Amazon, Spotify, etc. to stream music. Those accounts were paid for with real money, not stolen credit cards.
That streaming can generate a bigger royalty bill than the payments required to receive the content is a business model problem at the streaming companies.
[1] https://arstechnica.com/tech-policy/2020/03/court-violating-...
Now use this to play content for which you get royalties.
Ed: oops I see the charges have specific quotes about intent to evade detection
> For example, on or about December 26, 2018, SMITH emailed two coconspirators that, “We need to get a TON of songs fast to make this work around the anti-fraud policies these guys are all using now.”
The guy lied for financial gain. That makes it fraud. An argument about "defeating anti-fraud technology" is not the leg you want to stand on.
> For example, on or about December 26, 2018, SMITH emailed two coconspirators that, “We need to get a TON of songs fast to make this work around the anti-fraud policies these guys are all using now.
They are referring to anti-fraud policies as a general thing across all streaming platforms, not a specific named filter used by Spotify/Amazon/etc. Since they are referring to the policies by the intent rather than name or generically as ToS, this indicates an awareness that their behavior was ultimately fraudulent.
I understand it would be very different if the conspirators said “evade Spotify’s dumb ToS policies” and later Spotify said “actually those aren’t mere ToS, those are anti-fraud policies.” That’s not what happened.
In the legal sense of the term fraud detection is impossible, since these platfroms are not measuring the intentions of users.
Isn't lying for financial gain is what almost any company does?
Like what?
Would this have made any difference if a person was listening to those streams?
The fundamental "problem" for Spotify is that this arbitrage exists. It's unclear if any analogous behaviour would count as fraud.
Eg., if a taxi company pays $x/mile, and fares are $y/mile, and f(x - y) is "just right", I can drive around with an empty taxi and make money.
It "feels" like fraud because it's free money, which is basically what arbitrage is -- but in most cases we think its the market's problem to fix, not a legal one.
If this approach involved paying for lots of user accounts, using those accounts to stream music, uploading that music yourself, generating that music etc... then there shouldn't actually be a problem. You have paid for the songs you've listened to.
So spotify should have, in place, some system to stop users listening to more songs than their subscription covers. If they arent doing this, then the law isn't going to help. Their biz model has a clear iteration-to-zero in it.
It is clear fraud.
The answer is for Spotify to properly manage downloading so that people who download music are actually paying for it.
If you need 100 accounts to download 1,000,000 plays then whatever Spotify has to payout for those plays has to come from those 100 accounts. Whether those accounts are bots or actual people doesn't change the irrationality and going-to-zeroness in the business model.
EDIT: Unless they're using ad-funded free-to-play accounts, which is more of a problem. Specifying exactly what the fraud is here is difficult. They arent providing music deceptively, they arent fraudently downloading it -- it may be against TOS, but spotify isnt being sold anything in that download. If there's some sort of hybrid-way of talking about fraud here, its really across both actions (uploading knowing that users wont listen to ads?) and (downloading without listening to ads?).
It comes down to exactly what transaction terms we think are implied in the upload. You're not selling something to spotify. Likewise, when you download, you arent buying anything.
> at relevant times to this Indictment, a Manhattan-based music distribution company ("Distribution Company-1 ") [...] required customers, such as SMITH, to "agree not to engage in (or to permit, encourage, enlist, retain, or employ third parties to engage in) [...] any activity and/or method which involves the artificial creation, by human or non-human means, of online or offline plays on audio and/or audio-visual streaming services, where such plays do not represent bona fide end-user listening and/or views initiated by genuine consumers and taking place in the reporting country."
Moreover, the arbitrage available here doesn't really have anything to with "burning investor money". Aside from the fact that Spotify has been profitable so far in 2024, this fraud would be possible even if they had been profitable during the whole time the fraud had been carried out: it doesn't require that Spotify's royalty outlays be greater than their subscription/ad revenues in the aggregate. Rather, it relies on the fact that royalties are calculated in terms of total streaming volumes, but subscription prices are volume independent.
So while it's possible for a single individual to abuse the system and generate more royalties for themselves than what they pay in subscription revenues, it wouldn't work as a method in the aggregate for Spotify users to en masse produce royalties that exceeded streaming revenues. Even in unprofitable years their revenues greatly exceed their royalty payments. E.g., Spotify reportedly paid out $9B in royalties for 2023, on over $14B of revenue. At that point they were still operating at a loss due to R&D, marketing/sales, and administrative expenses, but that's all quite aside from their royalty calculation system.
def., Fraud, "knowing misrepresentation of the truth or concealment of a material fact to induce another to act to his or her detriment"
For it to be fraud, they would have had to relevantly misrepresent themselves in some transaction. The question is whether their actions were misrepresentations. This is unclear.
If I upload a song to Spotify I am not promising anything wrt to who listens to it or how. If I stream a song from Spotify I'm not promising to listen to ads, or only to listen to as many songs as are profitable for Spotify. So it's unclear that any misrepresentation has been made.
Spotify is well within its rights to terminate these accounts for TOS violations, and there might be some crime in taking payments from spotify.. but this is quite unclear.
There's a real problem in saying that whatever BS a company puts in its TOS defines fraud and if you dont do that you're defrauding a company. If that's true, then you're probably defrauding companies all the time, as are many.
Accepting the facts in the indictment at face value, I really don't see how this is unclear. The contract between Smith-the-artist and Spotify was that they would pay him royalties for legitimate streams which were not the result of manipulation, and that he would not deliberately manipulate the streams.
Smith then knowingly violated these terms, attempted to conceal his violations from Spotify, and continued to accept the royalty payments to which he knew he was not entitled. Aside from the misrepresentations required to initially set up the scheme (e.g., signing up bot accounts under fake names), he also repeatedly and directly denied participating in the exact sort of scheme he knew he was perpetuating: "I have never done anything to artificially inflate my streams". The allegations in paragraphs 26-30 cover several instances of this.
> If I upload a song to Spotify I am not promising anything wrt to who listens to it or how.
I'm not sure how you are squaring this with the quoted portion of the indictment from my prior comment. Smith did, in fact, make promises regarding who would listen (and how) to the songs he uploaded: namely, that he would not personally or via a third party listen to the songs in a manipulative manner.
> There's a real problem in saying that whatever BS a company puts in its TOS defines fraud and if you dont do that you're defrauding a company. If that's true, then you're probably defrauding companies all the time, as are many.
That's simply not the case here. Smith was not inadvertently running afoul of fine print buried deep in TOS that he never read to trivial detriment of some corporation. He was engaged in repeated and deliberate misrepresentation with the intent of inducing Spotify and distribution companies to pay out royalties in excess of $10M to which he knew he was not entitled.
If those were premium accounts, then I doubt it would be fraud.
That's certainly not the only misrepresentation Smith made throughout this scheme.
I'm having trouble understanding why you think it's particularly relevant whether these were free or paid accounts (as alleged in the indictment, the fraud made substantial use of paid family plan accounts).
It's worth noting that the scheme here was not merely bilateral between Spotify and Smith--it involved multiple streaming services, multiple distribution companies, the "Rights Organizations", and financial service providers who were all part of the overall scheme of delivering unearned royalties to Smith based on fraudulent streams.
If the facts alleged in the indictment are true, it seems immensely clear to me that Smith made multiple false representations to each of these parties as part of the scheme to have them deliver streaming royalties to which he knew he was not entitled. Among those false representations were: - false names provided on accounts for the purposes of concealing his identity - statements to financial service providers in order to obtain debit cards in those false names - agreement that he would refrain from engaging in manipulative streaming activity - insistence that he was not engaging in such manipulative activity when suspicion s arose
Smith directly responded to the suspicions that arose as early as 2018, but the scheme continued for years afterwards. So even if one were to make the case that he was initially unaware of a clause buried deep in the TOS, he was made expressly aware that he was in violation of those terms and chose to continue deceiving the parties (and indeed increased his efforts to conceal the fraudulent activity from the streaming platforms).
Now, it may be the Spotify alone wouldn't have a case for damages here since his activity didn't directly cause them to pay out more total royalties, and his subscriptions increased the total pool. But this isn't a civil case where Spotify is suing for breach of contract to claw back the royalty payments; there was $10M in royalties paid out to Smith that came from somewhere which was owed to someone other than Smith. Smith lied to multiple parties involved in the music steaming business in order to induce them to make those royalty payments.
I'm no lawyer, and so don't have a great idea if there's technical elements to the charges here that will be difficult to prove beyond a reasonable doubt, but I think it's more than clear that this was not 'fair play' on the part of Smith. I am reminded of the viral 'free money glitch' that went viral several weeks ago on TikTok: people were writing fraudulent checks, depositing them at ATMs and then withdrawing the cash portion that Chase permits you to withdraw before the check has fully cleared. That the automated systems for preventing this sort of abuse were imperfect does not somehow make it not fraud.
> SMITH spread his automated streams across thousands of songs to avoid anomalous streaming as to any single song. SMITH was aware that if, for example, a single song was streamed one billion times, it would raise suspicions at the Streaming Platforms and the music distribution companies that those streams were the result of streaming manipulation. 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 much smaller number of times. As a result, SMITH repeatedly identified the need for more songs as crucial for facilitating the fraud scheme. For example, on or about December 26, 2018, SMITH emailed two coconspirators that, “We need to get a TON of songs fast to make this work around the anti-fraud policies these guys are all using now.”
What kind of AI generating music tech were they likely to have been using in 2018?
Techno demoscene style music. It's amazing what you can do with just 4kb.
You can listen to it on Spotify
https://open.spotify.com/album/1TwLLLOArGizLfbzR2Pbgi
https://en.wikipedia.org/wiki/Aleatoric_music?wprov=sfla1
But even crazier is so did Mozart
https://en.wikipedia.org/wiki/Musikalisches_W%C3%BCrfelspiel...
So at the end of cases like this if they can recover the "damage" cause by the fraud will the artists get a compensation based on recalculated stats or the streaming company just pockets the money? I know in cases of ad click frauds sometimes Google compensates but that's a little different and I'm curious if there was cases for this on streaming platforms.
The fraudulent behaviour is on the part of the streamers. There should be no cross user account royalty dependecy.
There are 3 ways you can go about paying artists.
Way 1 is to say that x% of what you pay goes to the artists and that the streaming service takes the rest, no matter how many songs you play.
This makes no sense; this would cause a Taylor Swift fan that listens to her for hours on end, but also listens to other music sometimes, to pay less to Swift than somebody who just plays one Swift song per month and nothing else.
Way 2 is to pay a set amount per play, but make sure that the amount is low enough that if you play music for 24 hours a day, 7 days a week, the payouts are still less than what you paid for the service.
Most people don't listen to music for that long, so this makes artist payouts unrealistically low.
Way 3, which is what services actually do, is to have a much higher per-play rate and assume that users listen to n songs a month on average.
There will be some users who listen to a lot more, and you will take the loss on them, but by the law of large numbers, you will make an overall profit.
This is predicated on the fact that nobody "cheats" the service and misrepresents how much music they've been listening to. If you fraudulently tell Spotify that you've listened to your own songs 10m times this month, you make a lot of money.
What's your source of this? I thought that all money is pooled, and then divided by total play statistics - i.e. actually numbers, not assumed values based on some guessing in advance as you suggest. That would make their business super risky. Imagine there's a holiday season and everyone listens double the usual amount. That could bankrupt a streaming company. Also, their incentives would be to discourage users of playing music, and we don't see any evidence of it.
> Way 1 is to say that x% of what you pay goes to the artists and that the streaming service takes the rest, no matter how many songs you play.
> This makes no sense; this would cause a Taylor Swift fan that listens to her for hours on end, but also listens to other music sometimes, to pay less to Swift than somebody who just plays one Swift song per month and nothing else.
This makes total sense for me as a consumer - I want my money to go exactly to the artists I'm listening to. I don't want any of my money to go to Taylor Swift or any other artist I don't care about. There are several benefits to this model:
- it would benefit small niche artists who have couple hundreds devoted fans. Maybe it would make it possible for some of them to live off the streaming, instead of getting pennies. It would reduce payouts to super starts, but I don't care about that since I'm the one who's paying and I vastly prefer that my money goes to who I want
- it would be inherently more fair than the existing model
- it would absolutely eliminate any kind of fraud as described here. You won't be able to invest $100 and receive $200 by manipulating numbers, you could only receive $100 - streaming fees
> This makes no sense; this would cause a Taylor Swift fan that listens to her for hours on end, but also listens to other music sometimes, to pay less to Swift than somebody who just plays one Swift song per month and nothing else.
If I'm understanding the first quoted paragraph correctly, the second is not necessarily the result.
Say I'm paying $10/month to the streaming service and they pay $7/month to the artists no matter how much or how little I play.
Whether or not the second paragraph holds depends on how they distribute that $7/month among artists. They could for example take that $7, divide it by the number of songs I listened to that month, and pay each artist I listened to that amount times the number of their songs I listened to. In that case the result would be that your described.
Another way they could decide the value of a song is by taking the total amount they have to distribute, $7 x the number of subscribers, divided by the total number of songs listened to, and pay each artist that times the number of plays of their songs. Let's call that amount per play a share. In that case a person who only listens to one song a month and that song is a Swift song causes Swift to get one more share than She would have if they had not played any Swift songs. A person who listens to many Swift songs in a month causes here to get many more shares than she would have if they had listened to no Swift songs. The results is not what is described in your second paragraph.
You could assign a pot per users and let the user distribute the pot according to some algorithm.
Assigning other users' pots to the suspect's bots seem like a rather bad algorithm.
If there was a per user pot, manipulation like the suspect did would give bogger all.
I guess the fundamental problem is that the streamers want to manipulate the royalties.
He was also coordinating the botnet streamers. From the first sentence of the first paragraph of TFA:
> a scheme to create hundreds of thousands of songs with artificial intelligence and use automated programs called “bots” to stream the AI-generated songs billions of times.
rightbyte, your laziness contributes nothing to this forum and it would be better if you deleted your account. Your comment was bad and you should feel bad.
My point is that I see the streaming payout scheme as fraudulent. It is a block box 'trust me bro'.
Clearly, it is possible to abuse it and the streamers want to be the ones doing it.
I understand perfectly well what the FBI and music industry think is the fraud. They wanted some other label to have the money.
The AI is not really relevant except that it made the crime harder to detect.
Where's the line between that and requiring every service that allows media uploads to build a YouTube-esque content ID system, that would cost north of €/$/£100m and still be inaccurate/gameable?
inb4 "extracting monetary value" - can you come up with a fair definition that still clearly and unambiguously allows e.g. an aspiring opera singer to upload a demo of their own performance with their CV/application?
20th century's ideas of copyright have so far scaled incredibly poorly into the digital age. 30 years ago nobody expected an average (but highly motivated) Jane/Joe to be able to afford recording an album OR launching a company literally from their bedroom, yet both are now everyday occurrences.
Had he created songs by hand they would rack up millions/billions of botted streams but them not being international hits anyone had ever heard before would have made it obvious botting behaviour.
I'd prefer if the "justice" department used its resources to go after those who launder intellectual property.
> at relevant times to this Indictment, a Manhattan-based music distribution company ("Distribution Company-1 ") [...] required customers, such as SMITH, to "agree not to engage in (or to permit, encourage, enlist, retain, or employ third parties to engage in) [...] any activity and/or method which involves the artificial creation, by human or non-human means, of online or offline plays on audio and/or audio-visual streaming services, where such plays do not represent bona fide end-user listening and/or views initiated by genuine consumers and taking place in the reporting country."
The issue here is not a TOS violation on the subscriber side. Smith had an obligation not to engage in streaming manipulation as an artist receiving royalty payments, and deliberately skirted that obligation (while attempting to conceal his activity) with the intent of inducing Spotify to pay him royalties which he was not legitimately owed. That's fraud.
Would love to see the money laundering explain. From the indictment it seems to claim that its inviolation of "Title 18, United States Code, Section 1956(a)(1)(B)(i)." Which states "to conceal or disguise the nature, the location, the source, the ownership, or the control of the proceeds of specified unlawful activity; or" [1].
So, receiving money from unlawful activity counts as money laundering? That seems to be at odds with my layman definition of it. Like it just seems like a double-jeopardy violation, you commit fraud for money and get dinged for both fraud and money laundering from the same crime?
There are legitimate reasons to use different bank accounts and legal entities to conduct business.
So the defendant needs to show that they had legitimate reasons for different accounts.