Voice cloning technology allows for the creation of a digital copy of a specific person’s voice and its use to generate new utterances. This is not merely another way to use a recording, as the voice is separated from the person and reproduced in the form of a model that can be used without the person’s further involvement.
A synthetic voice can utter words that were never spoken. It can be used in an advertisement, a phone call, or a recording published online. The person whose voice characteristics were replicated may be unaware of either the creation of the model or its subsequent use.
A dispute does not necessarily have to stem from the unlawful acquisition of a voice sample. The source recording may have been made with the consent of the person concerned, and the other party may have obtained the right to use that specific material. However, this does not automatically grant the right to use that recording as the basis for an AI model that will generate new utterances.
Similarly, consent to create a digital replica of a voice does not automatically imply consent to every subsequent use of that replica.
It would therefore be appropriate to distinguish between the right to use a specific recording, consent to the creation of a voice model, and the right to generate new content using it. These are separate issues, although they are often treated as if consent to the first action automatically encompasses the other two.
Is a voice a personal right?
Polish law does not provide for a separate right to a “digital voice.” Protection should be sought primarily in the provisions governing personal rights.
The basis for this is provided by Articles 23 and 24 of the Civil Code, and, with regard to monetary claims, also Article 448 of the Civil Code. The list of personal rights is open-ended. This allows for the protection of human characteristics that are closely linked to a person’s identity and enable their recognition.
With regard to the voice, the Supreme Court’s ruling of March 7, 2023, II CSKP 659/22, is of significant importance. The Supreme Court held that the voice is subject to independent protection as a personal right. It noted that, like one’s likeness, the voice constitutes an element of a person’s physical identity, and its individual characteristics allow for the identification of a specific person.
This ruling is of particular significance for cases involving AI. The Supreme Court also addressed the digital transformation of a voice, which leads to attributing a statement to a specific person that they did not actually make. Such interference may infringe upon the voice as a personal right, regardless of whether an existing recording was used or a new statement was generated.
Voice cloning changes the nature of this interference. When a traditional recording is disseminated, it still uses a statement that was actually spoken. A voice model, on the other hand, allows for the creation of new content. A person can be presented as the author of words they never spoke.
If the voice is used to attribute a specific position or participation in a project to someone, this may result in defamation. When the model is created using material from the private sphere, privacy protection also becomes a significant consideration.
A legal recording does not always mean that voice cloning is legal.
Situations in which the source material was obtained legally are much more complex than cases of obvious impersonation.
The actor may have recorded dialogue for a video game. The company may have obtained the right to use the recorded material. However, the mere fact of having legal access to the audio file does not imply the right to create a model capable of generating further utterances based on the recorded voice.
The scope of rights should be determined by the contract and the purpose for which the recording was made. Acquiring rights to specific material does not necessarily include the use of the voice’s characteristics to create a digital clone.
Nor can copyrights or related rights to a specific recording be equated with the right to the voice itself without further analysis. The subject matter of these rights is a specific intangible asset or its fixation. The voice, however, remains an element of a person’s identity.
If the agreement did not provide for the creation of an AI model based on the recorded voice, it is difficult to assume that the general consent to use the recording also covered this form of exploitation. The dispute may concern both an exceeding of the scope of the agreement and an unlawful infringement of personal rights.
For these reasons, consent to voice cloning should be regulated separately. The agreement should specify the purpose of creating the model, the duration of its use, and the group of entities authorized to use that voice. The possibility of generating content for other projects and the consequences of terminating the collaboration require separate provisions.
Consent to the creation of a clone does not automatically cover every manner of its use.
A separate issue arises in situations where a person has indeed knowingly consented to the creation of a digital replica of their own voice; however, the mere fact of giving such consent does not determine the permissibility of every subsequent use of the model.
A model may be created for the purposes of a specific production or campaign. Its subsequent use by another entity or for a different purpose may not be covered by the original agreement, which is why the terms of the contract are so important. Consent to the creation of the model itself is one thing; consent to make it available to third parties is another. The possibility of creating additional materials after the collaboration ends may also require separate provisions.
Not every instance of exceeding the scope of consent will lead to identical claims. In one case, a breach of contract may be the key issue; in another, an infringement of personal rights.
Lehrman v. Lovo: a dispute over the use of recordings to create AI voices
The issue of the scope of initial consent arose in the case of Lehrman v. Lovo, Inc., heard before the U.S. District Court for the Southern District of New York.
According to the plaintiffs’ allegations, the recordings were obtained through the Fiverr platform for limited projects and were subsequently used to create commercial AI voices. The plaintiffs claimed that voices resembling their own were offered by Lovo under different names.
The case has not yet resulted in a judgment finding the company liable. In a ruling dated July 10, 2025, the court partially granted Lovo’s motion to dismiss the complaint but allowed certain claims to proceed. Among other things, the allegations regarding breach of contract and commercial use of the voice under New York state law remained. However, some of the claims based on copyright and trademark law were dismissed or limited.
This procedural ruling clearly illustrates the difficulty in choosing a legal basis. Copyright law may protect a specific recording, but it does not automatically grant exclusive rights to the characteristics of a voice. Provisions protecting a person’s identity and the terms of the agreement under which the recording was originally made may be of greater significance.
U.S. law, meanwhile, is developing separate instruments for the protection of the voice. In Tennessee, the ELVIS Act has been in effect since July 1, 2024; it explicitly protects the voice and provides for liability for certain instances of its unauthorized use At the federal level, work is also underway on the NO FAKES Act.
When a cloned voice is used for fraud.
Voice cloning is not used exclusively in content production. A synthetic voice can be used to impersonate a person authorized to issue instructions or to extort money.
In such cases, the protection of personal rights is only one aspect of the analysis. At the same time, there may be grounds for criminal liability and claims for damages related to the harm caused by the fraud.
How should one respond upon discovering the unlawful use of a voice?
The first steps should be aimed at securing evidence. The material may be quickly removed, and the account from which it was published may cease to exist.
You should preserve the file itself as well as data that can help determine the location and time of publication. If the material was used commercially, the method of its distribution may also be significant. It is advisable to secure information indicating that listeners recognized a specific person in the synthetic voice.
Further action depends on the nature of the infringement. It is possible to issue a cease-and-desist order and demand the removal of published materials. In certain cases, the claim may also include a prohibition on further use of the model or sharing it with others.
In the event of an ongoing infringement, a motion for preliminary relief may be crucial. It allows for an attempt to stop the continued use of the synthetic voice even before the trial concludes.
The evidence will decide the dispute
In many cases, the greatest difficulty will not be identifying the infringed right, but determining who created the model and how.
The mere similarity of a voice may require an expert evaluation. Previous recordings provided to the defendant and the circumstances of their use may be relevant.
In a dispute involving a technology platform, logs may prove crucial. They allow for verification of which account was used to create the model and when this occurred. They may also indicate what materials were uploaded and who subsequently used the voice.
Statements made during the creation of the model may also be helpful, particularly those regarding ownership of the rights to the recordings, the consent of the person whose voice was replicated, or verification of the owner.
voice. Such information may later reveal what the user declared to the platform and whether they acted contrary to those declarations.
Therefore, the response to a violation should not be limited to requesting the removal of the material. While the prompt removal of the recording may be necessary, steps must also be taken in parallel to secure the data needed to identify those responsible for creating and using the synthetic voice.
The user, the client, or the platform?
Liability for the unlawful use of a synthetic voice requires determining the actual involvement of the various parties.
The most direct liability lies with the person who created the clone without the required consent and used it for a specific purpose. The situation becomes more complicated when the model was commissioned by a company and the technical implementation was entrusted to an external entity.
Simply commissioning the work does not relieve the client of liability if the client was aware of how the voice would be used or decided to use the model themselves. On the other hand, the contractor will not always be able to rely on the client’s assurance that they hold the required rights. The assessment depends on the circumstances of the specific case.
A platform providing the technology is in a different situation than the direct perpetrator of the infringement. It cannot be held liable for every instance in which a user employs the tool. The situation must be analyzed differently when the platform was unaware of a specific instance of misuse, as opposed to a case where it received a credible report and failed to take appropriate action.
In the European Union, the DSA is relevant, particularly the “notice-and-action” mechanism provided for in Article 16. To the extent that a given service is of a hosting nature, a precise report may give rise to an obligation to respond appropriately to specific illegal content. The DSA does not, however, impose a general obligation to monitor all user content.
The AI Act, however, will not replace civil remedies.
Effective August 2, 2026, the transparency obligations set forth in Article 50 of the AI Act will take effect — obligations that extend to providers based outside the EU as well, a point we explored in more detail when looking at whether the AI Act applies to U.S. companies. In the case of systems that generate synthetic audio content, the provider will be required to ensure that the system’s output is labeled in a machine-readable format. A person or entity using an AI system to create or modify audio material constituting a deepfake will, as a general rule, be required to disclose that the content was artificially generated or manipulated.
A synthetic voice may fall under the definition of a deepfake if the generated or manipulated audio material resembles an existing person and could be mistaken for authentic.
However, the obligation to label such material does not resolve the dispute over consent. A recording may be correctly labeled as AI-generated and still infringe upon the personal rights of the person whose voice was reproduced.
The AI Act therefore adds a layer of transparency obligations. Claims related to the loss of control over one’s own voice will continue to be assessed primarily under civil law and contractual provisions.
How can you protect your voice from unauthorized use?
In cases involving voice cloning, it is crucial to quickly secure evidence, determine the source of the model, and assess whether the use of the voice fell within the scope of the consent granted.
Our law firm assists clients in matters related to voice protection, personal rights, and the use of AI, both in responding to infringements and in drafting agreements that mitigate risks associated with synthetic audio content.
