AI voice cloning: what protects an artist

On 5 October 2026, Giorgia Meloni submitted an application to the European Union Intellectual Property Office to register as a sound mark a four-second recording in which she pronounces "Io sono Giorgia Meloni" twice. The application, number 019431219, covers certain products and services in classes 9, 41, and 45, including downloadable multimedia content, publications, cultural activities, and the organisation of political events. It remains under examination: no trademark has been granted as of yet.

The initiative has been explained as an additional tool against deepfakes. The case holds interest beyond politics. It raises a question that affects any actor, singer, voice-over artist or dubbing actor: what protects an individual when an artificial intelligence can make them say, with a voice recognisable as their own, phrases they never uttered.

What a sound mark protects

A sound mark protects a sign capable of identifying the commercial origin of specific products or services. It does not register a voice in the abstract. In this instance, the requested sign is a specific recording, with specific words, a sequence, and a particular sound production.

EUIPO guidelines establish that the comparison between sound marks is based primarily on the auditory impression. When the sound contains words, their content, rhythm, and sequence may carry more weight than the similarity of the voice or intonation. Two distinct phrases spoken with the same voice are not necessarily similar for the purposes of a trademark.

If the application is granted, Meloni will be able to invoke the mark against commercial uses of identical or similar signs for products or services within its scope, in accordance with the general rules of trademark law. This will not grant her general control over her timbre. A replica that makes her utter a different text may be recognisable to the public and, nevertheless, not be confusable with the registered sign. The mark may add a line of defence; it is not a general response to voice cloning.

A voice is not a work in itself

Public discourse often conflates six distinct levels:

The voice as a trait of identity.
A specific performance, whether sung, spoken, or acted.
The recording that fixes that performance.
The words or composition performed, which may belong to another person.
A human imitation of the voice.
A synthetic replica generated by artificial intelligence.

The natural voice, considered in isolation, is not a work protected by copyright. The text or composition may be protected; the performance generates its own rights for the performing artist; and the recording may be covered, among other layers, by the rights of the phonogram producer. Copying an existing recording and synthesising new phrases with a similar voice are distinct behaviours and may activate different rights.

In April 2023, Heart on My Sleeve circulated, a song with synthetic voices that mimicked Drake and The Weeknd. Platforms removed it following claims from Universal Music Group. There was no ruling on the merits, nor was the exact content of the claims made public. It was reported that the removal may have been based on the direct copying of a producer's sound signature. The episode demonstrated the problem precisely: the imitation of a vocal identity, the copying of a recording, and the potential use of works during training are not a single legal issue.

In Spain, the voice appears expressly in the law

Organic Law 1/1982, on honour, privacy, and one's own image, considers the "use of a person's name, voice, or image for advertising, commercial, or analogous purposes" to be an illegitimate intrusion.

The law also contains several relevant rules:

Consent must be express. It may be revoked at any time, although damages, losses, and justified expectations caused by the revocation must be compensated.
The memory of the deceased retains civil protection. The action corresponds to the person designated in a will or, failing that, to specific relatives. In the absence of these, the Public Prosecutor's Office may act as long as no more than eighty years have elapsed since the death.
Public interest may limit protection. The law generally excludes illegitimacy when a relevant historical, scientific, or cultural interest prevails. Application to a documentary, artistic, informative, or satirical work depends on the specific case and the balancing with the freedoms of expression and creation.

This path is not limited to proving that an original recording was copied. It may be relevant when an identifiable voice is used for advertising, commercial, or analogous purposes. But it does not resolve all scenarios on its own: non-commercial cloning, parody, an artistic work, fake news, or fraud may require the analysis of different norms and rights.

The law was drafted long before current synthetic replicas. I have not located a Spanish judgment that has applied its Article 7.6 to a voice clone generated by artificial intelligence. For this reason, it is advisable to distinguish the text of the norm from the response that courts may give to a case that is still new.

When resemblance is enough

Scarlett Johansson reported in May 2024 that the "Sky" voice in ChatGPT resembled hers after she had declined to participate in the product. OpenAI stated that it belonged to another actress, hired before contacting Johansson, and denied any intention to imitate her. The company suspended the use of the voice. There was no ruling determining copying or infringement.

The case raises whether protection should depend on demonstrating the technical copying of a recording or on the public identifying a person.

Tennessee responded in 2024 with the ELVIS Act, which expanded state protection of the right of publicity to include the voice and defined it as a sound identifiable and attributable to a person, whether real or simulated. It is an important reference for artists and music professionals, but its scope is territorial and subject to exceptions and limits linked, among other factors, to freedom of expression.

Consenting to a recording is not consenting to a replica

For those who work with their voice, the contract is a central preventive tool. It does not replace the norms that may apply nor does it guarantee, on its own, a rapid defence against third parties, but it allows for the delimitation of what the recipient of a recording may do.

SAG-AFTRA summarises its safeguards in three principles: clear consent, fair compensation, and control over the performance. Its agreements on digital replicas require specifying the uses and negotiating the applicable conditions. Some also provide for additional consent for new uses, the right to reject proposals, or the withdrawal of the replica upon leaving a platform.

The lesson can be transferred to any cultural contract. Authorising the recording of a performance should not be interpreted, without a specific clause, as authorisation to:

Create a digital replica of the voice.
Generate new phrases or performances.
Dub into other languages while maintaining vocal identity.
Use the replica in another work, product, or campaign.
Train, adapt, or evaluate a model.
Retain the replica once the commission is finished.
Transfer it to third parties.
Exploit it after death.

The agreement should separate each use and specify its purpose, authorised works or campaigns, duration, territory, languages, media, possibility of sub-licensing, security measures, right of review, remuneration, withdrawal, and the destination of the files upon termination of the relationship.

Labelling is not authorising

Article 50 of the European Artificial Intelligence Regulation establishes transparency obligations for certain artificially generated or manipulated content. Anyone deploying a system that generates or manipulates image, audio, or video constituting a deepfake must disclose that the content has been artificially generated or manipulated. In works that are evidently artistic, creative, satirical, or fictional, the obligation is adapted so as not to hinder the exhibition or enjoyment of the work.

Transparency allows one to know that a voice is synthetic. It does not determine whether consent existed for its creation, whether the recordings were used lawfully, or whether its dissemination infringes upon personality rights, intellectual property, data protection, or other regulations. A deepfake identified as such may still be unlawful.

Italy added a specific criminal response in 2025. The new Article 612-quater punishes with one to five years of imprisonment anyone who causes unjust harm to a person by transferring, publishing, or disseminating, without their consent, images, videos, or voices falsified or altered by AI that are capable of inducing deception regarding their authenticity. The provision therefore requires more than the mere existence of a synthetic voice: dissemination, lack of consent, aptitude to deceive, and unjust harm.

Who can defend themselves

The problem is not only which rights exist, but who can enforce them. A Prime Minister may file for a trademark; a well-known actress may obtain immediate attention when reporting a similar voice; a record label may lodge a complaint with platforms. A dubbing actor, an emerging singer or a local voice-over artist may possess legal arguments and, at the same time, far fewer opportunities to detect the use, identify the responsible party, preserve evidence, and sustain a claim.

Inequality also exists within the contract. Those who negotiate individually with a production company, a platform, or a technology provider do not possess the same leverage as a trade union or a figure capable of withdrawing their consent. For this reason, the protection of the voice should not rest solely on each professional reading a clause more carefully: collective conditions, effective withdrawal procedures, and accessible avenues for redress are required.

While that framework is being developed, it is advisable to review today what is being authorised when entering a studio or providing a vocal sample. A generic clause regarding recording, adaptation, or future uses should not conceal the creation of a replica. The practical question is not only whether the company may retain the voice, but what it will be able to do with it subsequently.

About the open conversation

This text does not constitute legal advice. It distinguishes the voice as a feature of identity, the performance, the recording, and the synthetic replica because each layer activates distinct rights. If you work with your voice, in dubbing, music, or the performing arts, or in intellectual property law, and you wish to clarify a point or discuss this interpretation, write to me.

The author's relationship with the cases

There is no economic, professional, or collection-based relationship with the individuals, companies, and institutions cited.

AI voice cloning: what truly protects an artist