AI-generated vocals and artist likeness: the copyright questions no one has fully answered
As AI tools that clone or synthesise vocal performances become commercially available, the music industry is grappling with consent, attribution and compensation. We map the current legal landscape and what rights songwriters and performers currently have.
AI tools that can clone a specific artist's vocal timbre, generate entirely synthetic voices in any style, or replicate a performer's delivery have moved from research laboratories into commercially available products within the past two years. The legal framework governing what is and is not permitted is still being established through litigation, legislation and industry negotiation. This article maps the current state of the rights landscape and what protections performers and songwriters currently have.
What AI vocal tools can now do
The current generation of AI vocal tools broadly falls into three categories. Voice conversion tools take a recording of any voice and convert it to sound like a target voice — a trained model of the target singer's vocal characteristics. These are the tools most directly relevant to the likeness question: they can create a recording that sounds as if a specific artist sang a specific performance, without any involvement from that artist. Text-to-speech synthesis produces natural-sounding speech from text input, including some systems that can be trained on a target voice. Music-specific AI vocal generation produces full singing performances from text lyrics and style prompts, without requiring a specific voice model — these can sound generically "human" or be trained on a particular genre or style without targeting a specific artist.
The legal landscape: what rights currently apply
The copyright in a vocal performance is held by the performer under performer's rights (in the UK, under the Copyright, Designs and Patents Act 1988). However, performer's rights protect the specific recording of a performance — not the abstract characteristics of how a person's voice sounds. The sound of a voice is not copyrightable. This means that a model trained on thousands of recordings to replicate a vocal timbre is, under current UK and US law, operating in an area not clearly prohibited by copyright. The right of publicity (in the US, a state-law right protecting against commercial exploitation of a person's name, image or likeness) may provide a basis for claims where a specific, identifiable artist's voice is replicated for commercial purposes. In the EU, the AI Act (2024) introduces some transparency requirements for AI-generated content. No comprehensive legal framework specifically addressing AI voice cloning exists in the UK as of 2026.
Industry responses and emerging standards
The major labels (Universal, Sony, Warner) have issued statements opposing AI vocal cloning of their contracted artists and have issued takedown requests for AI-generated tracks that replicate specific artists' voices on streaming platforms. Several streaming platforms (Spotify, YouTube) have updated their policies to require disclosure of AI-generated content and to allow rights holders to request removal of content that violates their rights. The industry body IFPI and several collecting societies have called for legislation to extend performer's rights to cover voice characteristics used for commercial AI training. In parallel, some artists have publicly authorised AI tools to work with their voices under licensing agreements — creating a licensed market alongside the unlicensed landscape.
Frequently asked questions
For personal, non-commercial, non-distributed use, the legal risk is low. For commercial use — releasing a track commercially, earning money from it, using it in advertising — the risks are significant. Depending on jurisdiction, you may be infringing the artist's right of publicity (US), performer's rights in the specific recordings the model was trained on (UK/EU), and potentially passing off (making the public believe the track is the artist's own work). The safest approach is: do not release commercially any AI vocal content that replicates a specific identifiable living artist's voice without their permission.
This varies by territory. In the US, the Copyright Office has consistently held (most recently in 2023) that copyright requires human authorship and that purely AI-generated content is not eligible for copyright protection. In the UK, the Copyright, Designs and Patents Act 1988 provides some protection for computer-generated works (Section 9(3)), but its application to AI-generated music has not been fully tested in court. The practical implication: a track that is entirely AI-generated (no human creative input beyond prompting) may have no copyright protection and cannot be enforced against copying.
Document the infringing use (screenshots, recordings, URLs). If the content is on a streaming or social media platform, use the platform's rights violation reporting mechanism — most have specific processes for performer's rights claims. Contact a solicitor or entertainment lawyer with experience in intellectual property. In the UK, the Musicians' Union and Equity both provide member advice on performer's rights. The law in this area is developing rapidly and professional legal advice is essential for any significant commercial infringement.