What Is a Digital Replica? The New Rules for AI-Generated Voices, Faces and Performances

Practical Explainer | September 12, 2026

A famous actor finishes shooting a movie. Months later, the studio wants the character to deliver three new lines without bringing the actor back.

A brand wants an athlete to appear in hundreds of personalized advertisements without filming hundreds of commercials.

A game developer wants a character to generate new dialogue indefinitely in the performer’s voice.

A record producer wants a deceased singer to perform a song that did not exist during the singer’s lifetime.

Technically, versions of all four are becoming possible.

Legally, they are very different questions.

The common thread is the digital replica: a computer-generated version of a real person’s voice or appearance that can make that person seem to perform without actually doing so.

And that is why a sentence buried in a contract about “digital likenesses,” “synthetic performances” or “AI-generated uses” may turn out to be one of the most valuable provisions in the entire agreement.

A digital replica is generally a highly realistic, computer-generated representation of an identifiable person’s voice or visual likeness.

Modern statutes increasingly define the term to cover two situations:

California, for example, uses essentially that formulation in its digital-replica laws. Its definition excludes ordinary authorized activities such as reproduction, sampling, remixing, mastering and digital remastering.

That distinction matters.

Removing a microphone from a shot is not the same thing as generating an actor delivering a new scene. Cleaning up an audio recording is not the same thing as making a singer perform a new song. Correcting an athlete’s lighting in an advertisement is not the same thing as generating that athlete endorsing a different product.

The law is beginning to draw lines between editing a performance and creating a performance.

The easiest way to understand the issue is to stop thinking about AI as a special effect.

Think of the replica as a substitute performer.

Suppose a company has enough recordings of an actor to create a convincing digital version of her.

The company can now potentially make that version:

At that point, the economic question becomes obvious.

If the digital version can do the work, who controls the digital version?

That is the problem modern digital-replica rules are trying to solve.

This is increasingly ordinary.

An actor may undergo a body scan. A voice performer may record material sufficient to create a synthetic voice. An athlete may authorize technology that can generate localized advertising.

The existence of the replica is not necessarily the problem.

The important questions are:

What can it be used for? For how long? In which projects? Who gets paid? Can it be transferred? Can the person reject particular uses? What happens when the deal ends?

That is why modern agreements increasingly separate consent to create the replica from consent to use it.

SAG-AFTRA’s 2023 television and theatrical framework, for example, requires consent for creation of covered performer replicas and informed consent for their use, with compensation requirements in many circumstances. Its 2025 Commercials Contracts go further: performer consent is required before creation, informed consent is required before use, and the proposed use must be described with reasonable specificity.

The distinction is crucial:

“You may scan me” does not necessarily mean “you may make me perform anything forever.”

This is the classic deepfake problem.

Imagine a company generates a recognizable athlete endorsing a gambling app even though the athlete never agreed to the endorsement.

There is currently no single nationwide federal right governing every such digital replica.

Instead, potential claims may come from several places.

State right-of-publicity laws can protect the commercial value of a person’s identity. California, for example, prohibits certain knowing unauthorized commercial uses of another person’s name, voice, photograph or likeness.

And this principle predates generative AI. Decades ago, courts recognized that deliberately imitating a distinctive celebrity voice to sell a product could violate publicity rights. In Waits v. Frito-Lay, the Ninth Circuit upheld liability arising from a commercial using a singer who deliberately imitated Tom Waits.

AI did not invent the underlying problem.

It made imitation vastly cheaper, faster and more scalable.

This may become the most important battleground.

Suppose a performer signs a contract stating: Producer may use Performer’s name, image, likeness, voice and digital replica in all media, now known or later developed, in perpetuity.

Does that authorize the producer to create entirely new performances twenty years later?

Depending on the agreement and jurisdiction, maybe not.

California’s Labor Code now makes certain digital-replica provisions unenforceable when they permit a replica to perform work the individual otherwise would have performed, fail to reasonably describe the intended uses, and were negotiated without specified legal or union representation.

Notice what California did not do.

It did not prohibit digital-replica licensing.

It attacked a particular problem: obtaining sweeping rights through vague language from an inadequately represented performer.

That points toward where the market is heading.

The fight is increasingly not over whether replicas may be licensed. It is over how specifically the license must explain what the buyer can do.

Death does not necessarily put a likeness into the public domain.

California now specifically addresses unauthorized digital replicas of deceased personalities in expressive audiovisual works and sound recordings. Its statute generally requires permission for covered uses, while providing exceptions for areas including news, public affairs, sports, comment, criticism, scholarship, satire, parody, certain documentary or biographical portrayals, and incidental uses. California’s postmortem publicity right can extend for 70 years after death.

New York strengthened its own postmortem digital-replica protections in December 2025, generally requiring authorization for covered uses of deceased performers’ digital replicas while retaining specified expressive exceptions.

So the question is not simply:

“Is this person dead?”

It is:

Who controls the relevant postmortem rights, in which jurisdiction, and what exactly are you trying to create?

Consent rules cannot be separated from free-expression principles.

A rule giving every person absolute control over every depiction of themselves could collide quickly with journalism, documentary filmmaking, criticism, satire and parody.

That is why digital-replica statutes often contain expressive-use exceptions.

Tennessee’s ELVIS Act, for example, protects voice as well as name, photograph and likeness—including simulated voices—but contains protections for uses involving news, public affairs, sports, comment, criticism, scholarship, satire, parody and certain other expressive contexts.

California follows a similar approach for its postmortem digital-replica rules.

The difficult cases will live near the boundary.

A documentary using an obviously reconstructed historical scene is one thing.

A fabricated video presented so convincingly that viewers believe a living person actually made a statement may be another.

There is no single answer. Think of the law as layers.

That overlap matters because one piece of content can implicate several rights simultaneously.

An AI-generated singer, for example, may raise questions about the singer’s identity, the recordings used to build the replica, the composition being performed, contractual rights, platform rules and consumer confusion.

Owning one right does not automatically mean owning all the others.

This is one of the most common sources of confusion.

A record label may own the copyright in a sound recording.

A studio may own the copyright in a film.

A photographer may own the copyright in a photograph.

That does not necessarily mean those entities own an unlimited right to generate new performances by the person depicted or recorded.

Courts have long distinguished a person’s identity from the copyright in a particular recording of that identity. In Waits, for example, the Ninth Circuit explained that the claim concerned the singer’s voice and identity rather than infringement of a particular copyrighted recording.

The U.S. Copyright Office reached essentially the same broader policy conclusion in its major digital-replica study: existing laws leave gaps because replicas do not fit neatly within copyright, publicity, trademark, privacy or other existing doctrines. It recommended a separate federal digital-replica right.

A good digital-replica provision should answer much more than yes or no.

The meaningful questions include:

Creation. May the company scan, record or otherwise build the replica?

Permitted uses. Is it limited to one movie, one advertisement, one video game or one campaign?

New performances. Can the replica generate material the person never performed?

Modification. Can the company change age, body, accent, language, wardrobe or physical characteristics?

Approval. Must the individual approve individual uses?

Compensation. Is creation compensated? Is each use compensated? Are residuals, royalties or reuse payments triggered?

Term. Does authorization last six months, ten years or indefinitely?

Territory and media. Where and through which platforms can it be used?

Training. May recordings or replica data be used to train other AI systems?

Security. Who must protect the replica from theft or unauthorized cloning?

Transfer. Can the rights or replica files be sold to someone else?

Deletion. Must the underlying model or replica be destroyed when the license expires?

Death or incapacity. Do rights survive, and who exercises them?

These details are no longer contractual housekeeping.

They determine whether a person has licensed a particular performance—or created a potentially reusable digital version of themselves.

SAG-AFTRA’s 2025 Commercials Contracts illustrate the direction of travel. They require specific consent protections, compensation for covered generated performances, security measures and restrictions on retaining digital replicas after authorized periods.

Not necessarily.

Copyright may govern source material or particular recordings. The use of a person’s identity can raise separate publicity, contract, trademark and replica-right issues.

Not automatically.

Creation and use can be separate legal permissions.

Not necessarily.

Saying “AI-generated” may address deception, but it does not magically create permission. The pending federal NO FAKES proposal expressly states that an AI disclosure or disclaimer would not itself be a defense to unauthorized use.

Increasingly, no.

Traditional publicity claims often focus on commercially valuable identities, but newer proposals and some statutes protect individuals more broadly. Tennessee’s law, for example, is framed around an individual’s protected name, photograph, voice and likeness.

Too broad.

Parody and satire receive substantial expressive protection, but merely labeling something a parody does not resolve every case. Context, deception, commercial use, statutory language and First Amendment principles still matter.

Federal law already addresses certain particularly harmful synthetic imagery.

The TAKE IT DOWN Act, enacted in May 2025, covers nonconsensual intimate imagery, including qualifying AI-generated “digital forgeries.” Covered platforms must provide a notice process and remove qualifying material and known identical copies within 48 hours of a valid request; FTC enforcement of those platform obligations began in May 2026.

But that is not a general federal digital-likeness statute.

The broader proposal is the NO FAKES Act of 2026.

As of September 12, 2026, it has not become law. The Senate Judiciary Committee reported S.4591 in June 2026, and it remains on the Senate calendar.

The reported bill would create a federal right to authorize uses of an individual’s voice or visual likeness in digital replicas. It would make that right licensable, require written licenses containing reasonably specific descriptions of intended uses, establish liability and platform procedures, and preserve specified exceptions for news, documentaries, historical and biographical works, commentary, criticism, scholarship, satire, parody and other qualifying uses.

If enacted, it would dramatically change the current patchwork.

Until then, the patchwork remains.

For actors and musicians, the replica may become a separate asset from the underlying performance.

For athletes and influencers, endorsement agreements increasingly need to address whether brands may synthesize new advertisements instead of merely reusing approved photographs and footage.

For studios, labels, game companies and advertisers, securing broad traditional NIL rights may no longer be enough. Replica creation, generated performances and training rights increasingly deserve their own provisions.

For agents and lawyers, the negotiation is shifting from “Can you use my client’s likeness?” to “What precisely may you make my client appear to do?”

For AI companies and platforms, authorization, provenance and takedown systems are becoming part of the compliance architecture.

And for consumers, the hardest problem may ultimately be epistemic: when a famous person appears on a screen and speaks directly to you, will you know whether that person ever actually said those words?

The biggest development is the NO FAKES Act.

The reported 2026 bill would move digital-replica law toward a national framework built around authorization, specific licensing, remedies, expressive exceptions and platform procedures. But because it remains pending, state law continues to matter enormously.

Collective bargaining will matter too.

SAG-AFTRA’s agreements already demonstrate that unions can establish detailed consent-and-compensation rules faster than legislatures can create generally applicable statutes. Its 2025 video-game agreement, for example, contains consent and disclosure requirements for digital replicas and negotiated compensation structures for certain generated performances.

Expect the next generation of disputes to focus less on whether AI was involved and more on four questions:

What was created?

What did the person consent to?

What did the contract actually authorize?

And did the eventual use stay inside that permission?

A digital replica is not merely an AI-generated picture or voice clip.

It can function as a virtual version of a human being capable of performing new work.

That is what makes the issue commercially important.

The law is moving toward a basic principle: technology may make it possible to separate a performance from the performer, but that does not necessarily separate the performance from the performer’s rights.

For athletes, actors, musicians, creators and anyone else whose identity has value, the crucial contractual question is therefore becoming much more specific than:

“Can you use my likeness?”

It is:

“What are you allowed to make my likeness do?”

Not exactly. “Deepfake” is a broad technological and popular term. “Digital replica” increasingly has specific statutory or contractual definitions focused on highly realistic, identifiable representations of a real person’s voice or visual likeness.

It depends on the jurisdiction, purpose and circumstances. Commercial use may implicate publicity laws; specific states have enacted voice- and replica-focused rules; contractual rights may also apply. Fraudulent, intimate, election-related and other uses can trigger additional laws.

Rights can often be licensed, but the scope and enforceability of the license depend on the governing law and agreement. Some newer rules specifically favor written, reasonably specific authorization rather than blanket consent.

Not necessarily. Ownership of the footage and authorization to create new synthetic performances can be different rights.

Technically, yes. Legally, the answer depends heavily on postmortem publicity rights, contractual rights, the jurisdiction, the nature of the work and any expressive-use exception.