Voice Cloning Law UK: Who Owns Your Voice?
If someone uses a recording of your voice to create an AI version of you, what rights do you actually have?
It sounds like a question with a straightforward answer. According to legal counsel Mathilde Pavis, it does not, at least not in the UK.
In this episode of High Notes, Melissa Thom talks to Mathilde about the laws surrounding voice cloning, digital replicas, and AI, and what they mean for professional performers and anyone whose voice is recorded through their work. Their conversation moves from Zoom calls and voice notes to casting platforms, performer contracts, international regulation and the practical steps people can take now to protect themselves.
Mathilde’s central argument is not that AI voice technology should be stopped. She describes herself as neither anti-AI nor anti-cloning. The issue is whether people understand what they are agreeing to and can meaningfully control how their voice is used.
Why a recording of your voice is more valuable than it used to be
A voice recording once had fairly obvious boundaries. You could play it back, copy it, or convert it into another format, but it remained essentially the recording you originally made.
Mathilde explains that machine learning has changed that. Technology can now analyse a recording and identify patterns within it, separating vocal likeness from accent, age, style, prosody, expression, pauses, and other characteristics. That means a recording can become the raw material for something new. Someone else can effectively “drive” a version of your voice and create speech you never recorded.
That shift is what turns an everyday audio file into something potentially much more valuable. It also means the conversation is no longer relevant only to professional voice actors.
Your Zoom calls and voice notes matter too
Melissa deliberately begins outside the recording studio. BRAVA works with actors and voice professionals, but also with people who use their voice in business, so she asks what someone recording Teams meetings, Zoom calls, or workplace voice notes should consider.
Mathilde’s answer is that the law doesn't stop applying just because someone isn't a professional performer. Her first recommendation is to look at the agreements governing your work. Check employment contracts, client agreements, and other terms for language on intellectual property, personal data, the assignment or transfer of rights, and waivers. If reading an entire agreement feels overwhelming, she suggests starting by searching for words such as “rights”, “waive”, “assign”, and “transfer”.
The key question is whether material containing your voice can be repurposed to clone or recreate you, or used for AI training without further consent.
Her second recommendation is to look at the technology platforms you use. That could include Microsoft, Google, Zoom, Slack, Notion or other workplace services. Mathilde recommends checking whether your account lets you opt out of having content used for model training, and not assuming every service handles this the same way.
For businesses, the same shift creates responsibilities in the other direction. Mathilde argues that organisations should become explicit about how they use AI, what protections they have in place, and how they handle client or employee data. She sees this becoming part of ordinary digital due diligence rather than something unusual or confrontational.
For performers, the contract can become the weak point
For someone whose income depends directly on their voice, the stakes become higher.
Mathilde describes UK protection as a patchwork rather than a single clear right to your voice. Performers’ rights and data protection can offer protection in certain circumstances, but contracts can dramatically change the position.
That becomes particularly important because many standard agreements were not originally written with AI cloning in mind. Clauses assigning rights “in perpetuity” and “in all media” may have been created long before synthetic voices became commercially realistic, yet they can now raise questions nobody contemplated when the language was drafted.
Mathilde calls the contract an “Achilles heel” in the current system. Broad buyouts can potentially weaken protections that the law might otherwise give someone, even though neither party was necessarily thinking about voice cloning when the original wording was created.
Her advice is therefore practical: start reading what you agree to. Search contracts for “AI”, “machine learning”, “rights”, “assign” and “waive”. The goal isn't to reject AI automatically, but to ensure you have explicitly and knowingly agreed to cloning on terms you consider fair.
What happens to auditions uploaded to casting platforms?
The same issue arises with pay-to-play casting platforms, particularly because performers may have uploaded auditions and voice recordings to them for many years.
Unlike a directly negotiated client contract, platform terms are generally take-it-or-leave-it. Mathilde argues that platforms also helped normalise a culture of broad voice buyouts by prioritising buyer convenience. Once AI and the possibility of selling or repurposing voice data are added to that model, those old assumptions become much more problematic.
Her advice is to check whether platforms let you opt out of your data being used for training or secondary purposes, and to exercise those rights where possible.
More broadly, she suggests treating any platform business as a potential data business too. The service you knowingly signed up for may be only one part of its model; analysis and use of the information flowing through that platform may form another.
And even if you have been using a service for years, Mathilde doesn't accept that it's too late to act. She recommends opting out now, contacting support if there is no obvious setting and continuing to exercise whatever rights are available. AI systems need both new and old data, so preventing future use still matters.
Why Mathilde describes UK law as unfit for purpose
The underlying problem, Mathilde argues, is how difficult it currently is to say that a person simply owns or controls their voice.
Rather than a single statutory right clearly stating what permission is needed before someone can create a digital replica of you, UK protection comes from several areas of law, each designed for different circumstances.
That matters because modern AI does not merely reproduce an original performance. It can extract aspects of it and create a new simulation. Existing law developed around older ideas of recording and reproduction, and Mathilde argues that AI stretches that framework far beyond what it was designed to handle.
She contrasts that with jurisdictions that already had broader personality or publicity rights before AI arrived. Those laws protect aspects of an individual’s likeness rather than targeting one particular technology, which can make them easier to apply when a new form of replication appears.
Should the law protect digital replicas, or your voice itself?
That leads to an important distinction the episode discusses.
One approach would be to introduce a specific digital replica right. In simple terms, that would regulate the act of digitally cloning someone and specify what consent or permissions are needed.
The alternative is broader: a personality or likeness right. Rather than protecting against one particular technology, the law would recognise an individual’s voice, face or likeness as something over which that person has defined rights.
Mathilde sees advantages in the broader approach because it is more likely to survive future technological change. If legislation protects the individual rather than today's particular method of cloning them, lawmakers do not necessarily need to reopen the statute book every time the technology changes.
However, she is ultimately less concerned with what the protection is called than with whether something effective is introduced. In her view, existing UK protections leave too much uncertainty around mainstream and commercial uses of synthetic voices.
How does the UK compare with the US and Europe?
The international picture is complicated too.
Mathilde points to US states such as New York and California, where personality or publicity rights already provide a stronger foundation for protecting an individual’s likeness. She also highlights the strength that collective bargaining can give unionised performers, particularly through SAG-AFTRA agreements, which can address not only AI rights but contractual and economic protections.
Meanwhile, the EU's AI Act introduces transparency requirements for some AI-generated or manipulated content, but Mathilde explains that personality and likeness protection remains uneven across countries. Some European legal systems already have strong protections, while others are still considering how to respond.
Denmark is discussed as one country actively exploring stronger protection. What Mathilde particularly likes about the Danish approach discussed in the episode is building on a legal framework people already understand, rather than inventing an entirely unfamiliar type of right.
She also stresses, however, that simply creating a right is not enough. If somebody can immediately sign it away through an unfair contract, the practical protection may be very limited. Any reform therefore needs to consider contractual bargaining power as well as the right itself.
What would meaningful protection actually look like?
Melissa asks Mathilde what she would want the UK Government to do.
Her answer is strikingly straightforward. Before cloning someone, they should be asked for written consent. That consent should not simply disappear inside a broad contract. The person should have enough information to understand what they are agreeing to and, where relevant, unions should be able to negotiate minimum rates and protections.
She would also like individuals to have clear rights to information about what data is collected, how long it is kept and where it will be used. The same should apply to the resulting digital replica itself. Someone should know where it exists, understand how it is used, ask for its deletion, and ensure its use is time-limited.
For Mathilde, the label attached to that right matters less than making it simple enough that an ordinary person can understand and use it without needing an expensive legal battle.
It is not too late to protect your voice
One of the most useful moments in the conversation comes when Melissa asks whether the horse has already bolted.
Mathilde’s answer is no.
She describes herself as a “realist optimist” and argues that people can still regain agency. The simplest place to begin is by asking questions.
When you are offered a job, ask whether AI is involved. If it is, ask how. Will your performance be modified? Will a replica be created? Who will use it? What exactly are you agreeing to?
Mathilde argues that the industry still lacks settled language for some of these conversations, which makes being proactive particularly important. Asking does not automatically mean rejecting a job. The answers may reveal that the proposed use is entirely acceptable to you. But without asking, you cannot make that judgement.
Decide where your own AI boundaries are
For performers, Mathilde recommends developing a personal compass around AI rather than treating every use as identical.
You may be comfortable with some forms of AI-assisted production and uncomfortable with others. Your acceptable category may be broad and your no-go area very narrow, or the reverse. What matters is understanding your own position before negotiations begin.
Ideally, those conversations should happen before the contract arrives. By the time the paperwork arrives, production may be moving quickly, and there may be little room for a thoughtful discussion. Asking early about AI use is therefore not, in Mathilde’s view, being difficult. It is part of behaving professionally and understanding the work you are being asked to do.
If you agree to a digital replica, define exactly what you are selling
Mathilde does not advise performers never to license digital replicas. Instead, she recommends project-specific, time-bound agreements.
The project should be clearly defined, the permitted use should be clear, and the duration should be stated. That gives the performer an opportunity to price what they are actually handing over.
She strongly advises against an unrestricted digital replica licence in perpetuity. In her words, that can amount to handing over “the keys of the kingdom”. A voice replica could potentially be passed on or used in ways far beyond the original job.
Before agreeing to an AI-specific contract, Mathilde recommends understanding who is creating the replica and what technology they are using; who will have access to it; what project it can be used for; how long the agreement lasts; what happens when it ends; whether you can require the replica to be deleted; and whether anybody has the right to sell or transfer it onwards.
A replica controlled by a professional production team for a defined piece of work differs significantly from an interactive version of your voice available for members of the public to prompt freely. The potential risk to the performer is very different.
AI riders, agents and the importance of getting it in writing
For agented work, Mathilde believes AI protection should form part of the agent’s job. If the client refuses to alter its terms, the agent should make the performer aware of the implications so they can make the final decision.
For work the performer negotiates directly, that responsibility inevitably falls back on the performer.
An AI rider can be useful, but Mathilde points out that not every voice job comes with a formal contract. Work may be agreed through email or even direct messages. In those cases, performers can still state their AI expectations in writing and include wording that makes clear AI use is not permitted unless separately agreed.
The important part is creating a record. As Mathilde puts it, “record is everything.” Without written evidence, it becomes much harder to establish what both sides intended.
Start by asking
After a conversation spanning technology, data, contracts, international law and regulation, Mathilde’s final practical advice is remarkably simple.
If you click “I agree” on an online service, find out what happens to your data and opt out of secondary uses where appropriate.
If you are negotiating a job, ask about AI at the beginning.
If the proposed use doesn't fit your values or professional boundaries, you can walk away.
None of those actions solves the wider regulatory problem. But they return some agency to the person whose voice is being recorded.
And despite spending much of her professional life examining the risks around synthetic voice, Mathilde remains optimistic about the future of the human voice. To her, the billions invested in recreating it show how much people value it.
Technology may increasingly reproduce a voice's sound, but she believes voice is more than sound alone. The person behind it, what they choose to say, when they say it, and why, still matters.
She suggests that may prove far harder to replicate.
Comments