AI Music Fraud: How Streaming Platforms Were Manipulated at Scale

AI-generated music streaming fraud showing automated bot plays and manipulated audio data on a laptop

AI-generated music and automated bot streams can manipulate digital platforms at scale. Image credit: KorishTech (AI-generated).

AI music fraud has emerged as a new form of digital manipulation, as a US man pleaded guilty to using AI-generated songs and bot streams to exploit music platforms. The Guardian reports that a US man pleaded guilty to defrauding major music streaming platforms using AI-generated songs and automated bot activity. The scheme involved uploading large volumes of AI-created tracks and artificially inflating their play counts to divert royalty payments.

This case highlights a structural shift in how digital systems can be manipulated. As seen in how AI systems are increasingly shaped by real-world data, the same technologies can also be used to exploit platform mechanics at scale.

How the Fraud Worked

The scheme relied on two components: content generation and stream manipulation.

First, large volumes of music were created using AI tools. Instead of producing a limited catalogue of songs, the system generated thousands of tracks quickly and at low cost.

Second, automated bot networks were used to simulate listeners. These bots streamed the tracks repeatedly, creating the appearance of real user activity across platforms such as Spotify, Apple Music, and YouTube Music.

Because streaming payouts are based on total play counts, this artificial activity redirected royalty revenue toward the fraudulent catalogue.

Over time, this resulted in more than $8 million in illicit earnings, supported by billions of fake streams.

The key issue is not the use of AI itself, but how it was applied. Creating AI-generated music is not illegal. The fraud occurred when automated systems were used to simulate real listeners and artificially inflate stream counts. Because royalty payments are based on genuine user activity, generating fake streams is treated as fraud, as it involves intentionally manipulating a payment system for financial gain.

Why AI Makes This Possible

AI music fraud does not happen automatically, but becomes possible when AI reduces the cost of content creation and enables automation at scale. In this case, the individual used AI tools to generate large volumes of music and combined this with automated systems that simulated user behaviour. The role of AI was to remove production constraints, while automation enabled the scale.

AI systems allow:

rapid generation of large volumes of music
consistent formatting suitable for streaming platforms
automation of repetitive processes

This removes the traditional constraints of time, skill, and production cost.

At the same time, bot systems can simulate listening behaviour at scale. Instead of manually inflating metrics, automated networks can generate continuous activity across multiple accounts.

This combination—cheap content creation and automated distribution—creates a system where scale becomes the main driver of impact.

Why Streaming Systems Are Vulnerable

Streaming platforms distribute revenue based on a pooled model. Total subscription revenue is divided among rights holders according to their share of total streams.

This creates a volume-driven system where:

more streams → higher share of revenue
artificial activity → direct financial impact

Detection systems typically look for irregular patterns at the level of individual tracks or accounts. However, when fraud is distributed across many tracks and accounts, it becomes harder to isolate.

As a result, large-scale automated activity can blend into overall platform behaviour.

Cases like this are typically identified through pattern analysis rather than individual anomalies. Platforms monitor unusual activity such as repetitive listening patterns, abnormal geographic clustering, or disproportionate stream-to-user ratios. In large-scale cases, these signals can trigger investigations that involve both platform analysis and law enforcement.

Traditional Fraud vs AI-Enabled Fraud

Fraud Type	Example Case	Content Creation	Scale	Detection
Traditional streaming fraud	Manual manipulation (pre-AI)	Limited catalogue, manual uploads	Low to moderate	Detectable via repeated patterns
AI-enabled fraud	Michael Smith case (2026)	Hundreds of thousands of AI-generated tracks	Extremely high (billions of streams)	Harder due to distributed activity

This comparison shows how AI changes not just the method, but the scale and detectability of fraud.

What Platforms Are Doing to Detect It

Streaming platforms have begun strengthening detection systems to address these risks.

Current measures include:

detecting abnormal listening behaviour (e.g., repetitive streams or unusual patterns)
identifying large-scale uploads of low-engagement tracks
using machine learning models to detect non-human listening behaviour
removing fraudulent content and withholding royalty payments

Some platforms also require disclosure of AI-generated content and apply additional checks to large-scale uploads.

This reflects a broader shift toward system-level verification, as seen in Microsoft’s approach to proving what is real and what is AI-generated online.

Other AI-Enabled Fraud Cases

This case is part of a broader pattern where AI tools are used to replicate or simulate human behaviour.

Examples include:

AI-generated voice scams used to impersonate individuals in financial fraud
deepfake video calls used to deceive employees into transferring funds

These cases follow a similar structure: AI generates convincing outputs, while automation increases the scale of impact.

My Take

This case shows that the impact of AI is not limited to improving systems, but also to exploiting them.

The underlying shift is not just the use of AI, but the scale at which it can operate. When both content creation and user behaviour can be automated, systems designed around human activity become vulnerable.

The key factor is not sophistication, but volume. AI enables large amounts of activity to be generated quickly and consistently, which can overwhelm detection systems that were designed for smaller-scale anomalies.

This suggests that platforms will need to adapt not only to new forms of content, but also to new patterns of behaviour. As detection systems improve, similar techniques may evolve in response, suggesting an ongoing cycle between automation and control.