Digital Replica Rights: Voice and Likeness in the NO FAKES Act Era

A convincing AI clone of someone's voice or face now costs a few dollars and a few minutes of source audio. The legal system that decides who is allowed to make that clone, sell it, or stop it, was built for a world where impersonation required a lookalike actor and a soundstage. That gap is what "digital replica rights" are meant to close — and it's why a piece of federal legislation most people have never heard of just cleared a major hurdle in Washington.

What a digital replica actually is

A digital replica is a computer-generated or computer-altered depiction of a real person's voice, face, or body that is realistic enough to be mistaken for an authentic recording or performance. That covers a wide range of technical approaches:

• Voice cloning — synthetic speech trained on a person's recorded voice, capable of saying anything typed into a text box.
• Face-swapped or fully synthetic video — a real person's likeness mapped onto footage they never appeared in.
• Digital doubles — 3D-scanned or AI-modeled versions of an actor's body used for stunts, de-aging, or posthumous performances.
• Style clones — outputs trained to imitate a specific performer's cadence, mannerisms, or delivery closely enough to read as that individual, even without using their literal biometric data.

Digital replica rights are the legal claim a person has over these reproductions — the ability to decide whether their voice or face can be synthesized, and to be compensated or to object when it happens without consent. They sit at the intersection of several older legal doctrines: right of publicity (a person's control over commercial use of their identity), copyright (which protects creative works but traditionally not a person's voice or face itself), and, increasingly, purpose-built statutes written specifically for AI-generated media.

Why existing law wasn't built for this

Historically, someone impersonating a celebrity's voice needed real talent and real effort — a soundalike singer, an impressionist, a body double. Courts developed right-of-publicity law around those cases: Bette Midler successfully sued over a soundalike jingle in the 1980s, and Tom Waits won a similar case in the 1990s. But right of publicity is a state-law patchwork, not a federal right. Protections, remedies, and even the definition of "identity" vary significantly from California to New York to Tennessee to states with no meaningful statute at all. Some states protect only living people; others extend rights to a person's estate after death, but for wildly different terms. None of them were drafted with a tool that can generate a stranger's voice from thirty seconds of audio in mind.

That mismatch is the reason there's been a rush, over the past two years, to pass laws that explicitly name AI-generated voice and likeness as a category — Tennessee's ELVIS Act being the most publicized example, alongside a slower parallel effort to establish a single federal standard.

The state patchwork that built up first

Before any federal bill got traction, states moved on their own, and the resulting landscape is a good illustration of why a single national standard is attractive to almost everyone touching this space — creators, platforms, and AI companies alike.

Tennessee's ELVIS Act, named with the state's most famous musical export in mind, was among the first laws to explicitly add "voice" as a protected category alongside name, image, and likeness, and to specifically contemplate AI-generated imitations rather than just human impersonators. Other states have extended existing right-of-publicity statutes to reference AI or "synthetic" media without a full rewrite, which leaves open questions about how courts will interpret decades-old statutory language against a technology the drafters never anticipated. A handful of states still have no meaningful likeness statute at all, meaning a resident's only recourse against an unauthorized AI clone may be a general tort claim like false light or intentional infliction of emotional distress — doctrines that were never designed for this and often require proving harm that's hard to quantify.

This patchwork creates practical headaches beyond the obvious "which law applies" question. A voice actor based in a strong-protection state can still have their clone generated by a company headquartered in a weak-protection state, distributed on a platform incorporated in a third state, and viewed by an audience everywhere. Choice-of-law fights in these cases are expensive and slow, and by the time a court sorts out jurisdiction, the unauthorized content has often already achieved whatever reach or damage it was going to achieve. That's the specific inefficiency a federal right is designed to remove — not by making state law irrelevant, but by giving every affected person a baseline claim that doesn't depend on geography.

Why it matters right now

In June 2026, the Senate Judiciary Committee unanimously advanced the NO FAKES Act — short for the Nurture Originals, Foster Art, and Keep Entertainment Safe Act — moving it a step closer to a full Senate vote. The bill would create something that doesn't currently exist in U.S. law: a licensable federal property right in a person's voice and visual likeness, enforceable against unauthorized digital replicas regardless of which state the person lives in or where the infringing content was made or hosted.

That "licensable" detail is the structurally important part. A right that can be licensed is a right that can be bought, sold, bundled into a contract, and enforced through the same mechanisms that already protect copyrighted works — including notice-and-takedown obligations for platforms that host infringing replicas. Unanimous committee advancement doesn't guarantee passage, but bipartisan, unopposed movement out of committee is a meaningful signal after several prior versions of federal likeness legislation stalled without a floor vote. For an industry that has spent two years operating on a state-by-state patchwork, a single federal standard would change how contracts, insurance, and platform moderation policies get written.

How digital replica rights are meant to work

The mechanics that NO FAKES and similar state laws propose are closer to copyright than to traditional privacy law. A few core features recur across the proposals:

1. The right belongs to the individual, not a platform or employer by default. Studios, labels, and ad agencies can license it, but the underlying right originates with the person whose voice or face is being replicated.
2. It survives death for a defined term, transferable to an estate — relevant for the growing business of posthumous performances and "hologram" tours.
3. It requires a takedown process, modeled on the DMCA, so a person (or their representative) can demand removal of an unauthorized replica from a hosting platform.
4. It carves out exceptions for news reporting, commentary, satire, biographical works, and other First Amendment-protected uses — the hardest part of drafting any of these laws, since an overly broad right can chill legitimate documentary and parody work.
5. It applies regardless of technical method, meaning the law targets the realistic output, not the specific AI architecture that produced it.

How it differs from existing protections

The practical gap this fills: today, a person harmed by an unauthorized AI voice clone has to figure out which state's law applies, whether that state's statute even mentions AI, and whether their situation fits an old doctrine written for lookalikes and soundalikes rather than synthetic media. A federal, technology-specific right removes that guesswork.

Practical implications for businesses and builders

For any company touching voice, video, or likeness — AI product teams, media companies, advertisers, game studios — this isn't an abstract policy debate. It changes what "consent" needs to look like in a product flow.

• Voice AI and avatar platforms need real consent capture, not just terms-of-service boilerplate. A user uploading someone else's voice sample to a cloning tool is a liability surface, not a feature request, once a federal right with takedown teeth exists.
• Advertising and marketing teams using AI-generated spokespeople, synthetic voiceovers, or de-aged footage of talent need licensing language that explicitly covers digital replica rights — most legacy talent contracts predate the concept and don't.
• Platforms hosting user-generated content should expect a takedown obligation analogous to DMCA notice-and-takedown, which means building the intake, verification, and removal workflow before it's legally required, not after.
• HR and internal tooling teams building AI avatars of employees for training videos or internal comms need the same consent discipline as external-facing products — an employee's likeness right doesn't disappear because the use is internal.
• Insurers and legal teams are already starting to underwrite "synthetic media liability" as its own category, distinct from general media E&O coverage.

The businesses least exposed are the ones that treat digital replica consent the way they already treat data privacy consent: explicit, scoped, revocable, and documented — rather than something inferred from a checkbox no one read.

A practical consent checklist

Teams shipping any product that touches synthetic voice or likeness can get ahead of this by building — not bolting on after a complaint — a few concrete controls:

1. Positive, scoped consent at intake. Capture explicit permission for the specific use case (e.g., "internal training video," not a blanket "any use"), not an implied consent buried in a long terms-of-service document.
2. Source verification. Confirm the person uploading a voice or image sample has the right to do so — a manager, agency, or platform account holder isn't automatically authorized to consent on someone else's behalf.
3. Usage logging. Keep a record of what was generated, when, and under what consent grant, so a takedown or licensing dispute doesn't turn into a forensic exercise.
4. A revocation path. Consent that can be given should be able to be withdrawn — build the technical ability to stop future generation and, where feasible, remove existing outputs.
5. A designated takedown contact. Whether or not a formal DMCA-style process is legally required yet, having a clear, monitored channel for likeness complaints reduces both legal exposure and response time.

None of this is exotic; it's the same discipline privacy teams already apply to personal data. The difference is that voice and face data is uniquely hard to anonymize or de-identify after the fact, which raises the stakes on getting consent right at the point of collection rather than trying to fix it downstream.

Real limitations and open questions

None of this is settled, and a few genuine tensions remain unresolved even in the most developed drafts of the legislation.

• First Amendment friction. A right broad enough to stop malicious deepfakes can also be broad enough to threaten satire, biopics, and commentary. Every version of NO FAKES has had to narrow its exceptions language in response to pushback from documentary filmmakers, comedians, and news organizations.
• Federal preemption is messy. If a federal right passes, does it override stronger state protections, or set a floor? States like Tennessee that moved fast on their own AI-specific statutes have an interest in keeping their versions intact, and reconciling that with a new federal layer isn't trivial.
• Enforcement against foreign and anonymous actors. A U.S. federal right does little against a deepfake generated and hosted overseas, or distributed through channels that don't respond to takedown notices. The law can compel U.S. platforms to remove content; it can't compel a bad actor abroad to stop producing it.
• Proving unauthorized use at scale. Detection tooling for AI-generated voice and video is improving but is not reliable enough to be the sole basis for automated takedown at platform scale — which means human review bottlenecks are likely for the foreseeable future.
• Who owns a "style," not just a literal voiceprint? The hardest edge case is an AI output that sounds like a performer's general style and delivery without cloning their literal voice data. Current drafts lean toward protecting recognizable identity broadly, but this is the area most likely to generate early litigation and inconsistent rulings.

What to watch next

The near-term signals worth tracking aren't just the Senate floor vote. Watch how state legislatures respond if a federal bill moves — some may pause new AI-likeness bills to wait for federal clarity, others may accelerate to lock in stronger local protections before preemption questions get decided. Watch how major platforms and AI vendors update their terms of service and consent flows in anticipation, since companies with the resources to get ahead of compliance tend to move before a law is finalized, not after. And watch the licensing market itself: once a right is clearly licensable at federal scale, expect talent agencies, unions, and AI companies to start building standardized licensing frameworks the way the music industry built mechanical licensing around copyright — turning a legal protection into an actual commercial pipeline for consented, compensated digital replicas rather than just a defensive shield against unauthorized ones.

FAQ

What is the NO FAKES Act?

The NO FAKES Act is proposed U.S. federal legislation that would create a licensable property right in a person's voice and visual likeness, giving individuals a legal claim against unauthorized AI-generated digital replicas and a takedown mechanism similar to copyright's DMCA process. It advanced unanimously out of the Senate Judiciary Committee in June 2026.

Are digital replica rights the same as right of publicity?

They overlap but aren't identical. Right of publicity is an older, state-by-state legal doctrine covering commercial use of a person's identity generally; digital replica rights, as proposed federally, are specifically designed around AI-generated voice and likeness and would apply nationally rather than varying by state.

Is it currently legal to clone someone's voice with AI?

It depends heavily on the state and the use. Some states, like Tennessee under its ELVIS Act, explicitly regulate unauthorized AI voice cloning. In states without a specific statute, a person may have to rely on older doctrines like right of publicity or defamation, which weren't written with AI cloning tools in mind and can be harder to apply.

Does a digital replica right survive after someone dies?

Under most current proposals, including NO FAKES, yes — the right can pass to an estate for a defined period, which is directly relevant to posthumous performances, AI-generated "reunion" content, and hologram-style tours using deceased performers' likenesses.

How would platforms be required to respond to unauthorized deepfakes?

The leading proposals model enforcement on the DMCA's notice-and-takedown system: a rights holder (or their representative) sends a formal notice identifying the unauthorized replica, and the hosting platform must remove or disable access to it to maintain safe-harbor protection.

Does this stop parody, satire, or documentary use of someone's likeness?

That's the intent of the exceptions built into the legislation, but it's also the most contested part of the drafting process. Current versions carve out news reporting, commentary, criticism, and biographical works, though how narrowly or broadly courts interpret those carve-outs will only become clear once cases are litigated.

What should a business do now, before the law passes?

Start treating voice, face, and likeness data the way you already treat personal data under privacy law: get explicit, documented, scoped consent before training or generating a replica, and build a removal process for unauthorized use now rather than waiting for a legal mandate to force it.

Teams building products that touch AI-generated voice or likeness and want help navigating consent, licensing, and compliance design can reach out to Woyce Technologies.