Sony Music says it has submitted requests to online platforms for the removal of more than 260,000 AI-generated deepfakes that impersonate its artists, including Harry Styles, Adele and Michael Jackson. The company reportedly told the Financial Times that identifying and tackling unauthorised AI tracks is becoming increasingly difficult.
The reported total, covered by Music Business Worldwide, concerns alleged artist impersonation rather than AI-assisted music as a whole. It is a measure of Sony's removal requests, not a ruling that every track was unlawful.
Key Takeaways
- Sony Music says it has requested the removal of more than 260,000 AI deepfakes from platforms.
- The requests concern recordings said to imitate the voices or identities of Sony artists.
- AI vocal processing and synthetic voices are not automatically the issue; unapproved impersonation and misleading presentation are the immediate risk areas.
- Producers releasing AI-assisted work should document consent, be clear about credits and avoid marketing that implies an artist's involvement.
What has changed
Sony Music had previously disclosed lower totals for AI deepfake takedown requests. The new figure indicates that the volume of recordings it considers unauthorised impersonations has continued to grow rapidly.
There is no new platform rule, court judgment or blanket ban on generative music in the information reported here. Sony's position is an enforcement one: it is asking services to remove specific material that it says impersonates its artists.
The distinction matters. A producer using a generated vocal texture, speech synthesis or an AI-assisted writing tool is not necessarily making a deepfake. Risk increases when the finished recording is designed to make listeners believe a recognisable performer sang on it, approved it or is commercially connected to it.
Sony Music's own AI usage terms are one available indication of how a major label is addressing AI-related use of its material and artist assets. They should not be treated as a substitute for checking the terms of the particular tool, distributor or platform being used.
The practical release risks for producers
Voice cloning is often treated as a production shortcut: a guide vocal can be replaced with a voice model, or a track can be reworked into the apparent style of a famous singer. The second approach can create problems well beyond the sound of the record if the vocal, title, artwork, metadata or social clips suggest a named artist is involved.
For independent producers, the sensible working principle is straightforward: obtain explicit written permission before cloning, modelling or commercially presenting another person's voice or likeness. This applies just as much to collaborators, session vocalists and unsigned artists as it does to major-label acts.
Keep a clear record of what was agreed: the source recordings supplied, whether they could be used to train or create a voice model, permitted territories and formats, approval rights, credits and any royalty split. A vocalist agreeing to record a chorus does not automatically establish permission for their voice to be synthesised into future songs.
The music itself may also carry separate rights questions. Producers should distinguish between permission to use a voice, permission to use a recording, and permission to use underlying songwriting or sampled material. Our guide to DJ licensing and copyright offers useful context on the wider need to clear music before it is shared or commercially used.
Release metadata and marketing deserve the same care
Distributors and DSPs are often the first practical checkpoint. Avoid putting a famous artist's name in the artist field, track title, feature credit, cover art or promotional copy if that person did not participate. Even a phrase such as “new track in the voice of” can make the intended imitation explicit and increase the likelihood of a complaint or removal request.
That caution also applies to DJ edits, bootlegs and teaser clips. A 20-second social post can still present a misleading voice imitation, and a private test upload can become a problem if it is publicly accessible or passed to promoters, labels or playlists.
Producers using AI tools should also read the tool's commercial terms before release. Those terms may address ownership, training inputs, voice-model permissions, attribution and restrictions on impersonation. They do not remove the need to secure permission from the person whose voice or identity is being used.
What to watch next
Sony's reported figure does not tell us how platforms assessed each request, how many recordings were removed, or which services received them. It does, however, show that large rights holders are actively monitoring AI impersonation at scale.
For producers, the immediate takeaway is less about avoiding AI altogether and more about maintaining a clean consent trail. Use AI-assisted vocals transparently, credit real contributors accurately and keep artist names, likenesses and distinctive voices out of a release unless you have permission to use them. The related disputes around AI training data remain a separate issue, as covered in Product London's report on an AI music company accused of training on copyrighted tracks.