Voice Actor Rights and Residuals in the Age of AI Voice Cloning
Copyright law leaves voice actors defenseless against AI clones of their own voices.
Voice actors are watching an entire industry get built out of their vocal cords, and copyright law offers them almost nothing to stand on. The synthetic voice market has gone from novelty to standard production option in five years flat; studios, ad agencies, audiobook publishers, and game studios now treat AI clones as a line item, not a gimmick. Venture money has followed, with investment in voice AI growing sharply between 2024 and 2025. A legal framework that actually protects the person whose voice made the technology possible in the first place hasn't followed. Instead, there's a scramble: state statutes, union contracts, and consent clauses trying to do the work that copyright law was never built to do.
Working voice actors have put the anxiety plainly: there's no certainty the work being done now will still exist in five years. That's not paranoia. It's a rational response to a market where a handful of famous voices are getting paid handsomely to license themselves, while everyone else gets an ultimatum: sign the terms in front of you or lose the gig to someone who will. That gap, between actors with enough leverage to negotiate and actors who have none, is the fault line running under everything that follows.
What AI voice cloning actually does to a performer's voice and why it's so hard to detect or stop
Voice cloning works by pulling apart the pieces that make a voice recognizable: timbre, pacing, accent, the particular way someone leans on certain syllables. A model gets built from those pieces, and once it exists, it can generate sentences the original speaker never said, never recorded, and may never even hear until the finished product ships. A few minutes of sample audio is now enough for a usable clone in commercial settings; feed a model more material and it can produce output good enough for audiobooks and national ad campaigns. Platforms like ElevenLabs, a voice cloning and text-to-speech tool, sit at exactly this tier of the market.
The technical detail that matters most, legally, is this: the output is not a recording of the actor. It's a model derived from recordings, a mathematical stand-in that behaves like the voice without being the voice. Recent demonstrations of real-time conversational AI voice systems in gaming show where this is headed. Nobody recorded that as a fixed audio file played back on demand. It's a system generating speech on the fly, in response to whatever a player types. Once a voice becomes a live, responsive system rather than a captured performance, the legal frameworks built around recordings start to look like they're describing a different technology entirely.
What copyright law actually protects — and why a voice actor's voice isn't in that category
U.S. copyright law protects original expression fixed in a tangible medium: a specific take, a specific recording, a specific performance captured at a specific moment. It does not protect the underlying voice, the tone, the cadence, or the accent that made that performance sound like a particular person. Courts aren't wrestling with an ambiguous corner of doctrine here; it's settled, and has been for a long time.
The Lehrman v. Lovo case, decided in the Southern District of New York in July 2025, put a fine point on it. The court's handling of the federal trademark and copyright claims made clear that federal law simply does not reach the attributes of a voice itself. Plaintiffs asserting those claims were left standing on state law, because federal law had nothing to offer them. Compounding the problem, there's no federal right of publicity at all, no national statute covering name, image, or likeness the way copyright covers a screenplay or a song. So the hole isn't confined to copyright; it runs through the entire federal IP structure. A company can build a voice model from an actor's recordings without that actor's knowledge and, under current federal law, infringe nothing.
The patchwork of state right-of-publicity laws and where they create real protection
With federal law offering nothing, protection falls to the states, and the states don't agree with each other. Right of publicity, the legal doctrine covering commercial use of someone's name, image, and likeness without consent, exists in some form in roughly half the states. Coverage, scope, and available remedies differ from one statute to the next, which means a voice actor's legal standing can change entirely depending on which state they live in.
Some states, including California and New York, have more developed statutory frameworks than others. Tennessee went further, passing legislation explicitly extending right-of-publicity protection to AI voice cloning. Federal legislation has been proposed that would create a national right covering digital replicas of voice and likeness, though no such bill has become law. Until that changes, protection depends on geography, which is a strange way to decide whose voice counts as property.
The gap isn't only legal, it's also moral, and the two don't always line up. When family members of deceased performers have publicly objected to AI recreations of their loved ones' voices, there has often been no statute for them to invoke. Those objections are about dignity and grief, not commerce, and the law has nothing built for that. Meanwhile, any company operating across state lines, or training a model on recordings scraped from performers in a dozen different jurisdictions, can effectively shop for the weakest law in the bunch. That's not a bug in the patchwork. It's the patchwork working exactly as a patchwork does.
Why consent has become the operative legal battleground in the absence of a property right
With no property right in the voice itself, consent is what turns a voice into something a company can legally license. No consent, no lawful basis for cloning it commercially, at least in theory. In practice, consent gets buried in terms-of-service language that most people never read closely enough to catch what they're signing away.
That's exactly what happened in 2025, when a major platform's terms of service drew public backlash for granting a "perpetual, irrevocable, royalty-free, worldwide license" over voice recordings and anything derived from them, a grant that survived even after the user deleted their account. This overreach wasn't an isolated one. Broad terms-of-service language is the standard mechanism by which voice data quietly moves from a one-time recording into a permanent training set, often without the speaker realizing what happened until much later.
Real informed consent requires knowing four things: what the recording will train, how long the resulting model gets used, in what contexts it will appear, and whether there's any way to revoke it later. Most standard agreements answer maybe one of those questions, if that. SAG-AFTRA's Sound Recording Code, ratified in April 2024, sets the higher bar: "clear and conspicuous" consent, with specifics about intended use, required before any digital replication happens. That standard exists inside a union contract. Outside it, for non-union freelancers negotiating alone against a client's legal department, no such requirement exists. They sign whatever's put in front of them, because the alternative is not getting hired.
How SAG-AFTRA built its AI protection framework through strikes, waivers, and authorized licensing deals
The video game voice actors' strike ran for about a year, targeting major studios across the industry. It followed an extended period of negotiations that hadn't produced meaningful AI protections, so the union walked.
The resulting agreement, ratified by a strong majority of voting members, introduced protections that didn't exist in any prior contract, including new rules around residual payments for digital replicas and restrictions on how those replicas can be deployed — terms that had to be won at the bargaining table rather than granted voluntarily. Separately, a waiver targeting AI audio commercials requires informed consent before a digital voice replica gets created and sets terms governing specific uses of that replica. That waiver targeted a specific corner of the industry, commercial voiceover, where cloning had already become routine before anyone had negotiated terms for it.
On top of the strike-driven protections, the union built out an authorized marketplace structure, with licensing agreements covering platforms across the industry, letting members license their replicas on terms the union has already vetted. Several things remain unresolved: the contract has a fixed term, and what happens when it comes up for renegotiation is genuinely open. Residual structures outside the rerun context remain unsettled. And every one of these protections applies only to union members. Non-union performers, which is most of the working voice actor population, get none of it.
What residual and compensation models are actually being used — and what remains unresolved
Three compensation structures have emerged so far. A flat licensing fee pays for building the model once, a per-use or per-minute royalty pays out each time the model generates new output, and a revenue-share model ties payment to whatever the finished content actually earns. Authorized marketplaces let actors set license terms within a structured system, which gives them at least some pricing power, rather than none.
Audiobooks make the tension concrete. A human narrator gets paid for the hours spent in the booth, then may earn royalties on top of that as the book sells. A clone of that same narrator's voice can generate an entire audiobook for a fraction of the cost of a recording session, and whatever residual the original narrator sees depends entirely on the fine print of whatever license they signed years earlier, possibly before anyone involved understood what that license would eventually be used for. High-profile performers who have licensed their voices for AI use are cases where enough name recognition existed to negotiate favorable terms. A working voice actor without that recognition has no comparable leverage. The contract says what the buyer wrote into it, full stop.
Nobody has answered the open question yet: if a cloned voice narrates ten thousand audiobooks, what share of that revenue belongs to the actor whose voice it is, and who's tracking the number closely enough to pay it out accurately?
What voice actors and enterprises should actually look for in contracts and consent agreements today
For actors, four contract terms matter more than anything else in the document: scope (which platforms and formats the license covers), duration (does it expire, or does it run forever), revocability (can consent be pulled back, and does the model get deleted if it is), and residual triggers (which specific uses generate additional payment). Any agreement that doesn't spell out all four is asking for trust the industry hasn't earned yet. Where an actor has no interest in licensing cloning rights at all, an explicit prohibition on using the recording for AI training should be the default ask, not an optional upgrade.
Enterprises commissioning voice work face the mirror version of the same problem. Documented, informed consent covering all four elements above isn't a nice-to-have anymore; it's closer to a legal necessity, and increasingly a reputational one. Which state law applies depends on where the performer lives and where the content gets distributed, and California and Tennessee currently impose the strictest requirements. Working through platforms that have already negotiated union-compliant licensing structures cuts legal exposure considerably compared to drafting a bespoke agreement from scratch with an individual performer who has no union backing and no leverage to push back on unfavorable terms.
Where the legal and compensation landscape is likely to move over the next few years
Federal legislation is the single biggest variable still on the table. A national right of publicity covering digital voice replicas would erase the state-by-state patchwork overnight and give every performer the same baseline protection, regardless of where they happen to live. The NO FAKES Act shows Congress is at least paying attention, but technology-specific federal IP law has a long history of moving slowly, and there's no reason to expect this round to be faster. In the meantime, expect more states to follow Tennessee's lead rather than wait for Washington.
SAG-AFTRA's interactive media agreement will come up for renegotiation, and that moment will be the real test of whether the protections built in recent years hold, expand, or quietly erode. The Lehrman v. Lovo ruling may get appealed, or other courts may take up similar claims and rule differently; how that plays out will decide whether voice actors ever get a federal hook to hang a claim on, or stay limited to state law indefinitely.
The economics point in an uncomfortable direction on their own. As voice AI gets cheaper and better, the business case for licensing a real person's voice weakens, unless the law forces companies to do it anyway. That makes legislation, not market pressure, the more likely driver of fair compensation going forward. For voice actors, the contracts and consent standards being set right now are the ones that will define the market for years. This is the moment to know exactly what's actually protected, and what has to be negotiated because nothing is protecting it yet.
