The law struggles to keep pace with generative models that can reproduce identity indefinitely. Meaningful digital replica consent must be specific, limited, and revocable. We must distinguish between a recording and a living model of a person.
The ability to capture a human voice or face and reproduce it indefinitely has outpaced the legal frameworks designed to protect them. A standard release form, once a simple agreement for a specific recording, now often grants broad rights to train models that can generate new content without further input. This shift creates a fundamental asymmetry between the creator of the data and the owner of the model.
We need to treat the concept of a digital replica with the same rigour we apply to physical property. The current default in many industries treats likeness as a commodity to be extracted once and used forever. This approach ignores the unique nature of biometric data, which cannot be changed if compromised.
Meaningful consent requires specificity, time limits, and the right to revoke. Without these safeguards, the power to define who we are in the digital realm shifts from the individual to the organisation holding the model. The following analysis outlines the mechanisms required to restore balance.
What a digital replica is
A digital replica is not merely a high-resolution photograph or a clean audio recording. It is a trained model, a set of weights and parameters that can generate new instances of a person’s appearance or voice that never actually existed. This distinction matters because a recording is a static artefact of the past. A model is a generative engine capable of producing infinite variations in the future.
When an organisation captures your voice for a single line of dialogue, they are not just storing that audio file. They are often using it to learn the timbre, cadence, and emotional range of your speech. This process creates a synthetic version of you that can speak new words, adopt new accents, or perform tasks you never agreed to. The replica acts as a persistent shadow of your identity.
This capability changes the nature of the asset. A photograph can be copied, but the original remains fixed. A digital replica can be iterated upon, refined, and applied to new contexts. The value lies in the generative potential, not the source data. Understanding this mechanism is essential for evaluating any contract that involves biometric data.
Old releases, new capabilities
Standard industry releases were drafted for a world of linear production. A voice actor signed a form to record a commercial. The studio paid for that session and retained the right to broadcast that specific recording. The scope was clear, the duration was defined, and the usage was limited to the agreed medium.
Existing contracts often contain clauses that grant perpetual, irrevocable rights to use the performer’s likeness in any medium, now or in the future. These provisions predate generative models and do not explicitly mention artificial intelligence, yet organisations may argue they cover AI training. This legal ambiguity allows such claims, although newer agreements increasingly address AI explicitly.
The problem is that the scope of "any medium" becomes dangerously broad when applied to synthetic generation. A clause intended to cover a radio spot might be stretched to justify creating a deepfake video of the performer. The contract fails to distinguish between using a recording and using a model. This gap leaves performers vulnerable to uses that were unimaginable when they signed the paper.
Right of publicity and its gaps
The right of publicity protects individuals from the unauthorised commercial use of their identity. It is a legal doctrine that recognises a person’s interest in controlling how their name, image, and voice are used for profit. This right is grounded in the idea that identity is a form of property. However, the application of this right to digital replicas is inconsistent.
Courts have struggled to apply existing precedents to new technologies. Some jurisdictions require proof of actual harm or confusion. Others focus on the commercial value of the likeness. The lack of uniformity creates uncertainty for both creators and platforms. An action that is illegal in one region may be permissible in another.
The core issue is that the right of publicity was designed for static representations. It does not adequately address the dynamic nature of a trained model. A model can generate a likeness that is similar enough to cause confusion but distinct enough to avoid direct copying. This grey area allows organisations to exploit the spirit of the law while technically adhering to its letter.
Four properties of real consent
To protect identity in the age of generative AI, consent must meet four specific criteria. First, it must be specific to the use case. A blanket permission to use your likeness for all purposes is insufficient. The contract must detail exactly how the data will be used, including the type of model and the intended outputs.
Second, consent must be limited in time. Perpetual rights are incompatible with the evolving nature of technology. A licence should expire after a defined period or upon the completion of a specific project. This ensures that the value of the likeness is not locked away indefinitely.
Third, consent must be revocable. An individual must retain the right to withdraw permission at any time. This right must include the removal of their data from training sets and the deactivation of any generated replicas. Without revocability, consent is merely a waiver of rights.
Fourth, compensation must be tied to usage. A flat fee for a session does not account for the ongoing value of a digital replica. Payments should reflect the scale and duration of the synthetic use. This aligns the incentives of the creator with the value they provide.
Performers, creators and ordinary people
The risk of uncontrolled digital replicas extends beyond professional performers. Ordinary people are increasingly exposed to biometric data collection through social media, video calls, and public cameras. The same mechanisms that allow a studio to train a model on an actor’s voice can be used to clone anyone with an online presence.
This democratization of replication tools lowers the barrier to entry for bad actors. Malicious organisations can harvest data from public sources to create convincing impersonations. The scale of this threat is amplified by the ease of distribution on digital platforms. Protecting identity requires vigilance from all individuals, not just those in the spotlight.
The principle of the face is the password now highlights the vulnerability of biometric identifiers. Unlike a password, a face cannot be changed if it is compromised. Once a digital replica exists, it can be used to bypass security systems or deceive contacts. This makes the protection of biometric data a matter of personal security, not just intellectual property.
Replicas after death
The issue of digital replicas does not end with death. Many contracts include clauses that grant rights to use a person’s likeness in perpetuity, which extends beyond their lifetime. This raises ethical and legal questions about the dignity of the deceased and the rights of their heirs.
Some jurisdictions recognise post-mortem publicity rights, while others do not. The lack of clear standards creates confusion for estates and organisations. A deceased person’s likeness can be used to endorse products or create content that contradicts their known values. This can cause distress to family members and damage the reputation of the deceased.
The principle that anonymised is a verb not a state applies here as well. Simply removing a name does not prevent identification if the visual or vocal traits remain. The legal framework must address the specific risks of synthetic media, including the potential for misuse after death.
Questions people ask
Can someone use my voice for ai without permission?
In many cases, yes, if you have signed a broad contract. Standard releases often include clauses that allow for future uses, including synthetic generation. Without explicit restrictions, organisations may interpret these clauses to cover AI training. It is essential to review any agreement that involves your voice carefully.
What is a digital replica?
A digital replica is a trained model that can generate new instances of a person’s voice or appearance. It is not a simple recording but a generative system capable of producing infinite variations. This technology allows for the creation of synthetic content that mimics the original subject with high fidelity.
Is it legal to clone someone's voice?
The legality depends on jurisdiction and the specific circumstances. Many places require consent for commercial use of a person’s likeness. However, the laws are often outdated and do not explicitly address generative AI. This creates a legal grey area where cloning may be permissible under broad contract terms or impossible to prove as infringement.
Close
The current approach to likeness rights is inadequate for the age of generative AI. Contracts that grant perpetual, irrevocable rights to use a person’s identity fail to protect the individual. They treat human identity as a static asset rather than a dynamic and personal attribute.
We must demand contracts that are specific, time-limited, and revocable. Compensation must reflect the ongoing value of synthetic use. These changes are not just legal formalities; they are necessary safeguards for personal autonomy. The power to define who we are must remain with the individual.
The technology will continue to evolve. The legal and ethical frameworks must evolve with it. We cannot rely on old definitions to solve new problems. Meaningful consent is the only way to ensure that digital replicas serve people, not the other way around.