AI Music Has a New Problem: the Listener Can Be Fake Too

AI-generated music is usually discussed as a creative problem: can a machine write a song, imitate a voice, or produce something we would still call authorship? In 2026, a less romantic question is becoming harder to ignore: what happens when making songs becomes cheap enough to manufacture catalogues at industrial scale — and when part of the audience can be manufactured too?

In early October, the Financial Times reported on Sony Music’s escalating fight against fake content and streaming fraud. By the end of September 2026, the company had asked platforms to remove more than 260,000 unauthorized tracks, almost twice as many as only a few months earlier. Part of the problem is made of deepfakes of famous artists. Another part is less spectacular and potentially more disruptive: music produced in bulk and pushed through artificial streams.

This is where AI music stops being only a new creative tool and becomes a problem of platform economics.

AI changes the scale of music production

Recording a song traditionally requires time, skills, instruments, production and distribution. Music generators reduce a large part of those costs. They do not guarantee quality or originality, but they make it possible to produce far more tracks in far less time.

That change in scale changes incentives. If the marginal cost of another track moves toward zero, generating thousands of songs can become economically rational even when each of them attracts almost no genuine audience.

The issue is not simply that more songs exist. Streaming systems are built around an economy in which every play carries some value, however small. Multiply production and plays, and fractions of a royalty can become an industrial model.

Streaming fraud: when the listener is synthetic too

Streaming fraud predates generative AI. Bots and services selling artificial plays have been used for years to inflate numbers, chart positions and royalty payments. AI adds another piece to the machine: now the content itself can be automated as well.

The combination is powerful. A system can generate thousands of tracks, distribute them through aggregators and use bot networks to produce streams. At that point, it does not need to persuade a real audience. It only needs to evade anti-fraud systems long enough to collect part of the revenue.

Deezer has said that more than half of new tracks uploaded to its platform can now be AI-generated, while also reporting very high fraud rates among streams associated with this material. Numbers on that scale explain why a simple “AI-generated” label cannot solve the problem.

Musical deepfakes are a different kind of problem

An anonymous synthetic track and a song imitating a famous singer’s voice belong to the same technological family, but they create different forms of damage.

In the second case, identity becomes part of the product. Analysis of AI singers and cloned voices looks at what happens when listeners can be led to believe that a performance belongs to someone who never recorded it.

For labels and artists, the harm is not limited to royalties. A fake song can exploit reputation, style and the relationship with an audience. It can also contaminate search engines and recommendation systems, making it harder to distinguish an authentic catalogue from synthetic material.

Abundance turns music into a fight for visibility

Music platforms have always managed enormous catalogues. AI pushes abundance to another level. If the number of uploaded tracks grows much faster than the number of hours people can spend listening, production becomes cheaper while distribution becomes more valuable.

Playlists, recommendations and fraud detection move to the centre of the business. It is the logic of the attention economy: when content is practically infinite, the scarce resource is no longer the song. It is the listener’s time.

That can increase the power of platforms. Whoever controls ranking controls which tracks actually enter the market for attention and which remain technically available but culturally invisible. Mise N Abyme has already followed the same problem from another angle in our essay on Spotify’s generative bubble: personalization stops being merely a filter when the platform can begin producing the material it recommends.

The royalty problem is not only moral

Streaming remuneration systems distribute a finite pool of money according to specific rules. If a growing share of plays is occupied by artificial or fraudulent content, the issue becomes one of redistribution.

Platforms are experimenting with different models and more aggressive filters. Every solution has consequences. A permissive anti-fraud system leaves money on the table for abuse; an overly strict one can punish independent artists and small but legitimate catalogues.

The debate around AI music and copyright remains important, but it is not enough. A perfectly legal synthetic track can still be used inside a streaming-fraud operation. A platform may have to decide what to do with millions of AI-generated tracks that do not necessarily infringe copyright at all.

The real transformation is economic

AI music is often presented as a democratization of creativity, and there is truth in that: it lowers barriers for people who want to experiment. The same reduction in cost, however, also enables an automated form of industrial production that does not require a real audience.

That puts pressure on a distinction streaming had already weakened: music as a work versus music as a unit of content. If a track can be generated, distributed and played automatically, the entire chain can function without a recognizable author and without a human listener.

That may be the most important AI-music question of 2026. Not whether a machine can make a good song, but what happens to a market when both supply and part of demand can be automated.

Sources and references

• Financial Times, Sony Music steps up fight against streaming fraud, October 2026
• Spotify, measures against impersonation, spam and deceptive content