A fake song does not enter the room alone

A fake song wearing a real artist’s voice is not just a weird internet prank. It arrives with somebody else’s name, somebody else’s reputation and the possibility of landing beside legitimate work on the same screen. The listener has to play detective. The artist gets another mess to clean up. The platform still gets to call the whole arrangement frictionless.

Sony Music told the Financial Times that it had asked digital platforms to remove more than 260,000 generative-AI tracks impersonating its artists by the end of September. The company had reported more than 135,000 such requests at the end of March. That is at least 125,000 additional requests in six months—an enforcement queue growing faster than any responsible system should accept as normal.

Read the number correctly

The number is serious, but it needs a label of its own. It counts tracks Sony says it asked platforms to remove. It does not tell us how many were actually taken down, how quickly services responded, which platforms received the notices or how often the same audio returned through a different distributor. Music Business Worldwide, which reviewed the Financial Times report, made those limits explicit. A takedown request is evidence of an enforcement burden, not a victory receipt.

That distinction matters because giant numbers can perform toughness while hiding the part fans and working artists need to see. If 260,000 complaints go into a black box and the public cannot see response time, repeat-upload rate or final disposition, the industry has described the size of the mop without telling us whether anybody fixed the leak.

AI music is not one big bucket

There is a reason this conversation gets sloppy. A fully synthetic prompt-made track, an unauthorized voice clone, a human recording cleaned up with generative tools and a song pushed by fraudulent streams can all end up under the same lazy headline: AI music. They are not the same creative or ethical act. Treating them as one category makes it easier for bad actors to hide and easier for useful tools to get blamed for the costume somebody else put on fraud.

In July, a coalition that included IFPI, RIAA, A2IM, independent-label groups, the Recording Academy, SAG-AFTRA and the Human Artistry Campaign proposed two voluntary track-level labels: AI-Generated and AI-Assisted. The difference is plain enough to fit beside an explicit-content badge. One means generative systems made the whole or primary creative performance. The other means human creators made the recording while using generative tools for limited expressive elements. That is not a perfect taxonomy, but at least it respects the difference between a tool and an impersonation scheme.

The platform is not an innocent shelf

Streaming services love to describe themselves like gigantic neutral record stores: millions of choices, all waiting for the listener. But the shelf has software. It accepts delivery, matches names, builds artist pages, powers search, makes recommendations and pays money. When a counterfeit record lands under a real person’s identity, every one of those systems has participated in making the fake legible.

Deezer offers one concrete example of what platform-side action can look like. The company said that fully AI-generated tracks averaged about 90,000 daily uploads in June and topped half of new deliveries on peak days. It also says it detects and labels fully generated music, removes it from algorithmic and editorial recommendations and excludes fraudulent streams from royalty calculations. Those are company-reported figures and policies, not a universal audit of streaming. Still, they prove that “the volume is too high” cannot be the end of the conversation when a service has already built systems to classify the volume.

A label helps only if the identity is real

Transparency cannot stop at a little robot icon. A correctly labeled synthetic song can still be fraudulent if it borrows an artist’s name, voice or page without permission. The required information has to travel together: who delivered the recording, whose identity appears on it, whether the voice is authorized, whether generative AI shaped the primary performance and whether the distributor verified those claims before release.

The industry’s proposed chart principles point in the same direction. A coalition of labels said AI-developed recordings should qualify only when the service used is lawful and authorized, the track is substantially human-made and there are no stream-manipulation concerns. That framework is still a proposal, and labels should not get to grade their own homework forever. But it acknowledges the real issue: provenance, consent and manipulation belong in the same system.

Fans should not need forensic ears

Deezer says a 2025 Ipsos survey across eight countries found that 97 percent of participants could not reliably distinguish fully AI-generated music from human-made recordings in a blind test, while 80 percent wanted fully generated music clearly labeled. Take those as company-commissioned research, not divine law. The common-sense point survives: “you should have known it was fake” is not a serious consumer policy when the product is designed to sound convincing.

Music has always involved illusion. Singers double vocals. Producers tune notes. Samples pull one era into another. Nobody needs a purity test pretending the studio was ever a witness stand. What listeners do deserve is an honest answer about authorship and identity. Technology can change the sound without stealing the name.

Publish the receipts

The next useful announcement is not another giant takedown total. Platforms should publish response times, confirmed removals, repeat-upload rates and the share of impersonation complaints stopped before release. Distributors should disclose how they verify artist identity and what happens to accounts that repeatedly deliver counterfeits. Labels should explain whether artists can review claims made in their names. The public does not need private case files; it needs enough aggregate reporting to know whether enforcement is improving.

Sony’s tally says the current cleanup model is losing ground. Fine. Then stop treating takedown as the finish line. Put provenance where fans can see it. Make identity verification part of delivery. Block repeat offenders across distributor accounts instead of chasing one URL at a time. Give artists a fast, documented appeal lane. And when a fake gets removed, keep it from walking back through the side door in sunglasses. The future of music can use new tools. It cannot run on counterfeit trust.