Sony Music’s number lands like a thunderclap: more than 135,000 AI-generated “deepfake” tracks removed from streaming services after takedown requests. Not 135,000 streams. Not 135,000 suspicious accounts. 135,000 tracks—songs built to sound like real, famous artists.

The first instinct is to treat that figure as a referendum on taste, authenticity, or the future of creativity. Yet the larger crisis isn’t aesthetic. It’s financial. The real fight is over the plumbing of streaming: how money moves, how fraud hides, and how quickly the cost of producing “content” has collapsed.

Deepfake songs are easy to hear once you know what you’re listening for. The more dangerous problem is quieter: streams that never came from listeners at all—bot-driven activity that can distort charts, inflate totals, and dilute royalty pools for everyone else. That’s where AI becomes less a novelty and more an accelerant.

Sony Music’s claim, reported in March 2026 via BBC coverage cited by multiple outlets, was stark: the company said it requested removal of more than 135,000 AI-generated deepfake tracks that impersonated its artists on streaming services. TechRadar’s coverage points to Dennis Kooker, Sony Music’s President of Global Digital Business, describing reputational and commercial harms from impersonations—everything from tarnished artist identities to derailed release campaigns. The examples tied to the reporting include targets such as Harry Styles, Beyoncé, and Queen.

That figure matters because it puts scale on something labels have been warning about for years: impersonation is no longer a fringe prank. It’s industrial. When the cost of generating a convincing vocal clone approaches zero, the marginal cost of flooding platforms collapses too.

Still, readers deserve the nuance Sony’s headline number can obscure. The reported 135,000 refers to takedown requests/removals of impersonations, not a census of all AI-made music on streaming platforms. The figure also doesn’t necessarily describe a single service like Spotify or Apple Music; it’s reported broadly as “streaming services,” without a public platform-by-platform breakdown.

A deepfake that mimics a superstar isn’t just “fake art.” It can be commercial sabotage. Release campaigns are carefully timed; a convincing impostor track can confuse fans, pollute search results, or ride the attention around an official drop. Even when streaming services eventually remove the track, the damage can arrive first—fast, algorithmic, and difficult to unwind.

Public debate often lumps everything into “AI songs,” as though the only question is whether computers can write good choruses. Sony’s removals highlight a narrower—and more legally and ethically charged—category: impersonation deepfakes, designed to resemble a specific artist.

That distinction matters because the motivations differ:

- AI-created music can be experimental, clearly labeled, and made without mimicking anyone’s identity.
- Deepfake tracks trade directly on another person’s name, voice, and public trust.

Even if a platform wanted to welcome AI-assisted creativity, it would still need aggressive controls for impersonation. A service can host synthesizer music without pretending it’s a human band; it cannot credibly host a vocal clone of a chart-topping artist while insisting the ecosystem is safe for creators.

Impersonation doesn’t only harm megastars. The broader market suffers when fans start doubting whether a new upload is real. That skepticism can reduce engagement across the board—especially for emerging artists who rely on trust and momentum. When discovery becomes a minefield, the safest choice is listening to what you already know.

A second-order problem follows: identification becomes a burden shifted onto listeners and artists. Fans are asked to do forensic work; musicians are forced to police platforms for counterfeits.

Deepfakes are flashy because they sound like someone. Fraud is more corrosive because it rewires payouts. Most major streaming royalty systems still broadly function on a pooled, pro‑rata model: platform revenue goes into a pot, and the pot is distributed based on share of total streams.

That means fraudulent streams don’t need cultural relevance to do damage. They only need volume. If bots inflate the denominator—total streams across the service—then the percentage allocated to legitimate music shrinks. The effect can be hard to see in any single report. It shows up as a slow leak: artists and labels feel underpaid compared to real audience demand, without being able to point to a single culprit.

AI makes fraud cheaper on both sides of the equation:

1. Content creation costs drop: generating tracks becomes fast and inexpensive.
2. Catalog scale becomes the strategy: large batches of tracks can be uploaded to spread manipulation across many IDs.

Deezer has explicitly connected the surge of AI-generated music to fraud incentives. Reporting and platform statements also describe a tactic that fits the pro‑rata model: fraudsters distribute artificial activity across many tracks, avoiding anomalies that might be triggered by one song receiving implausible plays.

If the Sony story is about identity, the most clarifying fraud story is about money. In 2024, U.S. federal prosecutors charged Michael Smith, a North Carolina musician, alleging a scheme that used hundreds of thousands of AI-made songs and bots to stream them billions of times, generating more than $10 million in royalties over a reported 2017–2024 period. Forbes covered the case, and Bloomberg Law framed it as a moment that brings scrutiny to streaming inflation.

The indictment’s theory—bots + huge AI catalog + streaming payouts—should reset the conversation. The harm isn’t just that listeners might be fooled by synthetic music. The alleged harm is that royalty money meant for working musicians is redirected to whoever can manufacture the most artificial activity.

A naïve fraud model looks for one song with an impossible spike. Modern manipulation can look like ordinary, low-level engagement across thousands of tracks. When AI can generate those tracks in bulk, the catalog becomes camouflage.

That reality also complicates enforcement. Platforms can remove an account; fraudsters can return with a new distributor, new uploads, and a new batch of tracks. The fight starts to resemble spam control: an ongoing contest over identity, verification, and incentives.

Streaming services have begun to treat artificial streaming as both a financial threat and a product-quality problem. The most concrete example in the research is Spotify’s move toward explicit penalties.

Distributor support documentation (including TuneCore’s) describes Spotify imposing a €10 per track per month charge when a track is deemed to have high levels of artificial streaming, passed from Spotify to distributors and then to the account holder. Those materials also note that Spotify’s detection methods are proprietary, and that play counts and royalty reports may be adjusted when artificial streams are removed.

That approach signals a shift: platforms are trying to push costs back onto the supply chain—distributors and uploaders—rather than absorbing the burden internally.

Penalty systems deter obvious manipulation, but they also raise hard questions. Detection is proprietary. Account holders may face charges or clawbacks without clear visibility into the evidence. A small label or independent artist caught up in suspicious traffic—whether from a bad marketing vendor or a malicious actor—can be punished quickly, while appeals move slowly.

A fair system needs two things at once:

- Strong detection and fast action to prevent dilution of payouts.
- Transparent standards and meaningful recourse so enforcement doesn’t become arbitrary.

Trust infrastructure—verification, labeling, and accountable distribution—starts to look less like bureaucracy and more like the foundation of the market.

Sony’s takedown figure points to an obvious policy proposal: label AI material. TechRadar’s coverage underscores that labeling is emerging as a “critical challenge.” Yet labeling alone doesn’t solve impersonation or fraud.

A useful label has to answer at least three different questions:

1. Was AI used in creation? (A broad category that includes benign tools.)
2. Is the work impersonating a real artist? (A higher-risk category.)
3. Are the streams legitimate? (A behavior question, not a content question.)

A track could be fully human-made and still be the target of artificial streaming. Another could be AI-assisted but honestly presented. Another could be a deepfake designed to trick fans. A single “AI” tag collapses these scenarios into one bucket and invites the wrong enforcement.

Platforms and labels are drifting toward a more demanding concept: provenance—a chain of accountability for who uploaded what, on whose authority, and with what rights. That includes basic identity verification and distribution controls, but also systems that can flag suspicious patterns at the account level.

The uncomfortable truth is that music streaming grew up in an era optimized for frictionless uploads. The next era will reward systems that can say “yes” to legitimate creators quickly while saying “no” to the industrial bad actors just as fast.

The story isn’t “AI is ruining music.” The story is that the cost of faking music—and faking demand—has dropped faster than the industry’s ability to verify either. Sony’s 135,000 removals show impersonation at scale. The Michael Smith indictment shows alleged fraud at scale. Deezer’s warnings show how tightly AI volume can couple to fraudulent incentives. Spotify’s €10-per-track penalty shows platforms experimenting with deterrence.

For artists and rights holders, practical implications follow:

- Identity protection becomes ongoing work. Monitoring for impersonation is now as routine as monitoring for piracy once was.
- Royalty integrity becomes a collective-action problem. Fraud drains the pool; everyone pays unless systems catch it early.
- Distribution choices matter more. The supply chain—distributor policies, upload verification, enforcement responsiveness—affects risk.

For listeners, the takeaway is less moral panic and more consumer literacy. If a “new” track from a major artist appears without the normal promotion, credits, and official channels, skepticism is rational. That skepticism shouldn’t calcify into cynicism; it should push platforms toward clearer verification signals.

A functional streaming economy will likely look more like a financial system: not because music should be sterile, but because money attached to scale invites industrial abuse. The goal isn’t to stop AI tools. The goal is to stop counterfeit identity and counterfeit demand from setting the price of everyone else’s work.