AI-generated music: Sony claims it can identify copied artists

With just a few clicks, platforms like Suno or Udio now allow users to produce tracks inspired by specific styles, eras, or artists. But behind this apparent ease, a question remains: which works were actually used to train these models?

Faced with the increasing number of disputes between rights holders and AI companies, such as the recent case with Anthropic, Sony intends to regain control.

The Japanese group, through its research arm Sony AI, has reportedly developed a technology capable of identifying the original works that contributed to training a generative model. A breakthrough that could redefine the rules of the game in intellectual property…

A detailed analysis thanks to “neural fingerprinting”

Revealed in the journal Nikkei Asia, the system relies on an advanced technical approach, based on “neural fingerprinting”..

Specifically, melodies, harmonies, and rhythms are converted into digital representations that allow comparison of an AI-generated track to a vast catalog of existing works.

If an AI actor agrees to cooperate, Sony can directly access the model's training data to identify the works used. In case of rejection, the tool performs a post-production comparison between the generated track and known tracks to estimate the likely sources of inspiration.

But the ambition goes beyond simple plagiarism detection, because the technology would be capable of quantifying influences. For example, the tool could indicate that a track is 30% inspired by The Beatles or 10% by Queen.

Thus, such granularity would pave the way for a proportional calculation of revenues from AI-generated music…

Towards a new revenue-sharing model?

For Sony, the stakes are strategic, because the group owns numerous catalogs, including a significant portion of Michael Jackson's, and is therefore directly exposed to the rise of synthesized music.

In 2025, Sony Music had already requested the removal of more than 75,000 pieces of content that imitated its artists.

With this approach, the objective goes beyond simple protection; it is rather about putting authors back at the heart of the economic equation, by creating a A redistribution mechanism based on the actual contribution of each work to an AI-generated track. This is a logic that some partnerships, such as the one between Universal and Udio, have begun to outline. For now, it is still too early to know if AI companies will voluntarily integrate this type of system, but at a time when competition is focused on model performance, the integration of intellectual property safeguards could be perceived as a constraint.