Sony Music says it has asked streaming platforms to remove more than 260,000 AI-generated tracks impersonating its artists, nearly twice the 135,000 it had identified by the end of March. The company-supplied count, reported Sunday by the Financial Times, captures removal requests rather than confirmed deletions. Even with that limitation, the increase offers a stark measure of how quickly synthetic audio is colliding with music rights, fan trust and streaming economics.

The reported deepfakes imitate performers including Harry Styles, Adele, Britney Spears and Michael Jackson. They are distinct from transparently labeled AI-assisted recordings made with permission: the central allegation is that the tracks trade on recognizable voices or catalogs without authorization. An independent music report published Monday confirmed the new total and its scope while noting that platforms have not publicly disclosed how many requests resulted in removals.

A surge measured in takedown requests

Sony has not released a track-by-track list, the names of every platform contacted or a public audit of the counting method. That makes the 260,000 figure a corporate enforcement metric, not a complete census of AI music online. Duplicate uploads, altered versions of the same recording and requests still under review could affect what the number represents.

Still, the direction aligns with public platform data. Deezer reported in July that roughly 90,000 fully AI-generated tracks were arriving each day at the June peak, exceeding half of new daily deliveries. The company’s disclosure said AI tracks remained only 1% to 3% of total listening, showing that enormous upload volume does not translate into comparable audience demand.

Deezer also said it classified up to 85% of streams on fully AI-generated tracks in 2025 as fraudulent and removed those plays from its royalty pool. Those figures come from Deezer’s proprietary detector, which the company says has 99.8% accuracy and fewer than one false positive per 10,000 human-made tracks. Because the model and underlying catalog are not independently auditable, the numbers are best treated as platform measurements rather than universal industry rates.

Fraud and imitation are different problems

The industry is confronting at least two related but separate threats. Voice or style impersonation can mislead listeners and infringe copyright, trademark or publicity rights even if nobody manipulates play counts. Streaming fraud, by contrast, uses bots or coordinated accounts to manufacture plays and divert royalty money. A synthetic track can be involved in either practice, both or neither.

That distinction matters because broad bans on AI-generated music could sweep in legitimate experimentation. In July, the International Federation of the Phonographic Industry adopted chart rules allowing AI-assisted work when the service is authorized, the recording is substantially human-made and the activity raises no manipulation concerns. The framework excludes purely synthetic or unlawfully trained work while leaving room for licensed creative tools.

Sony publicly endorsed that line. Its July joint proposal with Universal, Warner and independent labels tied chart eligibility to lawful models, meaningful human authorship and disclosure. The same approach separates authorized production tools from replicas designed to exploit an artist’s identity.

Distributors move upstream

Removing tracks after they appear is expensive and reactive, so labels are pushing distributors to screen content before it reaches Spotify, Apple Music, Deezer and other services. Sony joined the Streaming Integrity Initiative in September alongside Universal, Warner, CD Baby, The Orchard and more than two dozen other companies.

The initiative asks distributors to verify customer identities and rights ownership, screen uploads for infringement and AI-related risks, act against repeat offenders, share intelligence where lawful and measure the performance of anti-fraud systems. Distributors are a strategic choke point because a small number of services can deliver vast catalogs to many platforms.

But a common pledge is not the same as a common technical standard. The initiative does not publish shared error rates, appeal timelines or a neutral process for artists whose music is wrongly flagged. Independent musicians may have fewer resources to challenge automated removals, making transparent evidence and meaningful appeals as important as fast detection.

Academic work suggests that the upload pipeline is easier to exploit than industry policies imply. A June research paper tested 11 independent distributors by creating and submitting AI-generated music. The researchers reported inconsistent rules and weak enforcement, while 93% of the AI music they observed on Spotify received few or no plays. That supports two conclusions at once: synthetic catalogs can be uploaded at scale, but most do not win organic audiences.

The research was not a platform-wide audit and has not yet established how much revenue moves through fraudulent AI tracks. Detection systems also face an arms race. Generators change, files can be edited to obscure telltale artifacts, and classifiers trained on known models may struggle with unfamiliar ones. A system can be highly accurate in a controlled test yet perform less reliably across compressed, remixed or noisy audio.

For listeners, labels and artists, the immediate change is likely to be more identity checks, AI labels and silent removals—not the disappearance of synthetic music. Sony’s reported tally shows that enforcement is scaling, but it does not demonstrate that platforms are winning. The decisive measures will be how many impersonations stay down, how much fraudulent money is recovered, and whether legitimate musicians can contest mistakes without being buried by the same automation meant to protect them.