AI, voice cloning, and the future of music royalties
When artificial intelligence blurs the line between inspiration and imitation, who owns the music that follows?
British singer Jorja Smith’s label, the Orchard, is reportedly claiming royalties for the viral TikTok track “I Run”, produced by Haven. The track uses AI to manipulate the producer’s vocals into a female tone, which the Orchard claims is reminiscent of Smith.
Haven insists that the original vocals were theirs and were altered using AI prompts to emulate “soulful vocal samples”. Despite the song hitting #11 on Spotify US and going viral on TikTok and Instagram, it was removed from streaming platforms following takedown notices from the Orchard, RIAA, and IFPI. This dispute spotlights a growing dilemma: When does AI-enhanced creativity become outright infringement – and who’s on the hook when it happens? The producer (Haven), the platforms, or the AI itself (Suno)?
The risk of voice cloning and misattribution is at the heart of the issue. Smith’s voice is highly recognisable, and even imperfect AI imitations could mislead the public into believing that she performed or endorsed the track. Without clear AI labelling, listeners may not distinguish between human and synthetic performances. How closely must AI vocals resemble an artist before they risk misleading audiences and triggering passing off and/or moral rights violations? Even if the melody differs, imitating distinctive elements such as tone, phrasing, or expression could constitute passing off, or, in the absence of a clear disclaimer, false attribution. What steps should be taken to ensure that creators are fairly compensated and protected when their voice or performance is used without their consent?
Royalties and licensing add another layer of complexity. Haven maintains that, since the track “I Run” was produced using a model that had allegedly been trained on Jorja Smith’s works without authorisation, it is entitled to a share of the song’s royalties. However, the extent of Haven’s entitlement to royalties is unclear. The liability for any potential royalty may lie more with the AI provider (Suno) rather than the track’s creators (Haven producers), particularly if copyrighted works were used to train the underlying model, as alleged (but not yet proven) here. In any case, from a legal perspective, the contribution of Smith’s recordings to the contested track may be considered insubstantial where the output itself does not replicate the lyrics or melody of the original work(s). Furthermore, the territorial nature of copyright law may restrict claims further (as demonstrated by the recent decision in Getty v Stability AI). This case therefore highlights the difficultly of tracing influence and quantifying contributions in the context of AI-assisted music. Calculating royalties calculation is also fraught with complexity.
Ultimately, the case exposes the broader uncertainty surrounding IP law in the era of AI. The differences between US fair use, and European text and data mining exceptions, coupled with the territorial nature of copyright laws and the practical difficulties of obtaining evidence of copying, demonstrate the challenges faced by existing IP frameworks in keeping pace. How should global AI tools navigate the complexities of multi-jurisdictional IP laws and AI regulations? What responsibilities do they have with regard to labelling content, obtaining permissions, and sharing revenues with creators? “I Run” is more than just a viral hit. It’s a glimpse into a world where creativity, technology, and law collide.
Look out for our upcoming CMS Law Now article, where we’ll explore the legal complexities in this case in detail.
