Sony Is Chasing 260,000 AI Deepfake Songs. It Calls the Fight 'an Uphill Struggle.'

Somewhere on the internet right now, a song "featuring" Adele is streaming. It sounds like her, right down to the phrasing she uses in the chorus. The problem: Adele never recorded it. A generative AI model built the vocal from samples of her voice, and somebody uploaded it to a streaming service under a name close enough to hers that casual listeners never notice.

On October 5, the Financial Times put a number on that scenario. Sony Music has asked digital platforms to remove more than 260,000 AI-generated tracks impersonating its artists, through the end of September, nearly double the roughly 135,000 takedown requests it had disclosed six months earlier.

The steepest curve in the music business

Sony has been publishing these figures long enough to reveal a trend, and the trend is brutal. In March 2025, the company told a UK government consultation on AI and copyright that it had already requested more than 75,000 removals. In March 2026, Dennis Kooker, Sony Music's president of global digital business and US sales, put the figure above 135,000 at the launch of the IFPI's annual Global Music Report. Six months later, it exceeded 260,000.

Music Business Worldwide did the arithmetic: the jump from April through September works out to more than 20,000 removal requests per month. That is not a cleanup operation winding down. It is an enforcement department filing takedowns faster than ever and still falling behind.

The fakes target some of the biggest catalogs on earth: reports listed Harry Styles, Adele, Britney Spears, Queen, and Michael Jackson, the names most lucrative to clone.

What a "deepfake track" actually is

Sony describes the targets as "generative AI deepfake tracks that imitate our artists' voices and likenesses without permission." In practice, that covers two related tricks that platforms treat differently.

The first is the straightforward voice clone: an AI model trained on an artist's recordings generates a brand-new vocal performance, sometimes a fake collaboration that never happened, sometimes a "new" song in the style of an artist who has not released anything in years. The BBC, reporting on Sony's figures in March 2026, described these as songs "falsely purporting to feature" artists from its roster.

The second trick is sneakier: the fake gets delivered to a real artist's profile through the same upload systems legitimate distributors use. Spotify separates the two tactics, unauthorized voice clones versus fraudulent delivery to another artist's profile. Sony's tally does not distinguish them, so the 260,000 is a blend of both.

"We will keep enforcing our artists' rights and expect platforms to act quickly and to prevent the same content from reappearing. But it's becoming an uphill struggle."

The quote matters because of who it is aimed at. Sony has also sued AI music generators Suno and Udio over the use of recordings to train their models, but this request goes to the platforms, and the key ask is not just removal but stay-down: stop the same material from being uploaded again.

The whack-a-mole problem nobody has solved

Here is the central frustration buried in Sony's disclosure. The company reports how many tracks it has asked platforms to remove. It did not say how many were actually removed, how quickly platforms acted, or even which services received the most requests. The Financial Times did not report those figures either.

That gap is the whole story in one missing statistic. A takedown notice removes one file from one service. The same track, slightly renamed, slightly re-rendered, can be uploaded again in minutes through automated distributor pipelines. Rights holders keep filing new notices for content they have already reported, which is exactly what Sony means when it says platforms must "prevent the same content from reappearing."

Platforms are trying different answers: Qobuz labels AI-generated tracks and keeps them out of editorial features, Spotify is adding AI-identifying features and permits vocal impersonation only with the artist's authorization, and Australia's ARIA changed chart rules so wholly AI-generated recordings are excluded while substantially human-made AI-assisted recordings can still qualify. None of these stops a fake from being uploaded in the first place.

The wider crackdown is just beginning

Sony's disclosure lands in the middle of the busiest week the AI music debate has seen. The US Copyright Office opened a public inquiry into music streaming fraud, with AI-generated tracks squarely in scope, and is asking whether the "dilution of the royalty revenue pool" should be treated as a formal policy problem. A day earlier, a federal judge sentenced Michael Smith to 18 months in prison for a scheme that used bots to stream AI-generated songs and siphon more than $8 million in royalties.

On the legal side, the proposed US NO FAKES Act would give a concrete definition to a "digital replica": a newly created, computer-generated, highly realistic representation that is readily identifiable as someone's voice or likeness. Italy's competition authority opened an investigation on October 7 into Suno's terms of service, examining contract clauses rather than training data. The lawsuits over whether AI music generators may train on copyrighted recordings, Suno and Udio were sued by the majors, are still working through the courts.

Each of these is a different battlefield: fraud law, contract law, copyright law, and platform policy. Sony's 260,000 takedowns live on the platform-policy battlefield, the one that moves fastest and satisfies fewest people.

What the number doesn't tell you

It is worth being precise about what 260,000 measures. It is Sony Music's enforcement output against one specific problem: unauthorized impersonation of its own artists. It does not measure all AI-generated music, much of which is original, legal, and unlabeled. It does not count fakes on Sony's rivals' rosters, independent artists who cannot afford a monitoring operation, or the growing share of AI uploads on services like Deezer that are legitimate if uninspired.

In other words, the number measures one company's capacity to find the fakes, not the size of the flood. The flood is bigger, and it is rising faster than any label's enforcement team. That is what an uphill struggle looks like: not a battle being lost, but a slope that keeps getting steeper.

For listeners, the practical advice is simple. If a track "featuring" a famous artist appears with no announcement and no presence on the artist's own channels, check whether it is real before you share it. The fake Adele does not care who listens; the real one's royalties depend on you telling them apart.

References

Music Business Worldwide, "Sony Music has now asked platforms to remove more than 260,000 AI deepfakes," October 5, 2026 (reporting figures supplied to the Financial Times). MusicTimes, "Sony Music AI Deepfake Takedowns Top 260,000, Nearly Doubling in Six Months," October 7, 2026. The Music Network, "Sony Music Seeks Removal of Over 260,000 AI Deepfake Tracks," October 2026. World Lifestyler / USA Newshour, Deezer AI upload statistics and Copyright Office inquiry coverage, October 2026.