Who Owns Your AI Likeness?

For most of the history of media, licensing a person meant hiring the person. A singer went into a studio. An actor went onto a set. A model appeared at a shoot. A voice actor recorded the lines. Artificial intelligence separates the performance from the performer: once a sufficiently capable digital representation exists, new material can potentially be generated without that person being physically present.

That technical shift creates a rights problem before it creates a creative one. The U.S. Copyright Office defines digital replicas broadly around digital technology that realistically reproduces an individual’s voice or appearance, and in 2024 it concluded that unauthorized digital replicas exposed gaps serious enough to warrant a new federal right. Its Copyright and Artificial Intelligence initiative has since become one of the clearest public references for the policy problem.

At the same time, unions and rights holders have been building the commercial rules for the authorized version of the same technology. SAG-AFTRA’s current AI resources distinguish digital replicas from wholly synthetic performers and emphasize clear consent and defined intended uses. Its 2026 interactive-media guidance says performer consent for digital replicas must be written, conspicuous and tied to a reasonably specific description of use. That is not merely a labor issue. It is the beginning of a licensing schema.

The key distinction is not real versus AI. It is authorized versus unauthorized.

Much of the public conversation about deepfakes treats synthetic media as a binary: authentic media on one side, fake media on the other. Licensing markets introduce a third category — an artificial performance that is synthetic and authorized.

A cloned voice can be used for fraud; the Federal Trade Commission has repeatedly warned about the harms of AI-enabled voice cloning and impersonation. But the same basic capability can also power a performer’s approved multilingual narration, a licensed game character, an estate-authorized historical voice, or a creator’s commercial digital twin. The technology alone does not tell a buyer which situation they are in.

That is why provenance matters. A useful licensing record needs to answer questions such as: Who claims the authority to offer this identity? What part of the person is available — voice, face, motion, likeness or a fuller digital replica? Does the person approve individual uses? Are there restrictions on politics, advertising, adult content, training, geography or term length? Is the price public, negotiated or revenue based? And when was the information last verified?

Those are the questions Likeness Index is designed to organize. The marketplace-source index tracks where authorized listings originate and keeps the source attached to each claim. A person appearing on several marketplaces should ultimately resolve to one dossier rather than several disconnected search results.

A new market is forming, but it is fragmented.

Voice licensing developed early because speech is relatively easy to capture, generate and distribute. Visual likenesses and full digital replicas are now following. Athlete name-image-likeness infrastructure is converging with AI permissions. Music rights organizations and performers are negotiating voice and persona protections. Motion capture creates another layer: a recognizable performance can include body, gesture and movement even when a face is not the central asset.

The result does not look like a single marketplace. It looks like dozens of specialized pools: voice libraries, actor rosters, estates, athlete NIL platforms, digital-twin agencies, creator products, music services and performance-capture companies. Some expose public catalogs. Some reveal talent only to verified buyers. Some have APIs. Others operate through private partner feeds. Some disclose pricing; many do not.

This fragmentation is why a neutral index can be useful. Likeness Index does not need every marketplace to behave the same way. It needs to preserve the differences. A buyer searching for a recognizable voice for a game should be able to compare authorized options without assuming that one marketplace’s terminology or consent process applies everywhere. The same principle applies to a brand searching for an athlete, an instrumentalist, a dancer or an estate-administered personality.

You can already browse the index by emerging use cases through the category guides or search the underlying records directly in Find talent.

Consent is becoming infrastructure.

A useful signal is how specific recent agreements have become. SAG-AFTRA’s AI contract resources do not treat consent as a permanent blank check. They repeatedly connect permission to the particular replica and intended use. That suggests a future in which a “digital likeness” is not one asset with one yes/no flag; it is a bundle of permissions that can change over time.

Technically, that points toward rights metadata that behaves more like software permissions than a headshot in a casting database. A person might permit narration but not political advertising; games but not model training; a defined campaign but not perpetual reuse. An estate might authorize a voice in one context while withholding a visual replica in another. A performer might revoke or renegotiate availability.

The U.S. Copyright Office’s digital-replica work is also a reminder that copyright does not neatly answer all of these questions. Identity, publicity, contract, labor and consumer-protection law overlap. The Office’s report recommended a distinct federal protection against unauthorized replicas precisely because existing doctrines leave gaps. For buyers, that complexity increases the value of knowing the provenance of a purported license instead of merely finding a technically convincing model.

The likely winner is not the biggest library. It is the clearest chain of permission.

Generative quality will continue to improve and, in many cases, converge. When several systems can produce a convincing human performance, the differentiator becomes less about whether the output sounds or looks real and more about whether it is usable: Who approved it? What can you do with it? Can you prove that approval? What happens when the rights change?

That is the thesis behind Likeness Index. We are not an agency and we license nothing. We are building a source-attributed map of the authorized market — one that can say not merely that an AI version of a person exists, but where the permission came from and where the buyer must go to obtain the rights.

The distinction matters because an index that treats every synthetic human as equivalent would simply make deepfakes easier to find. An index built around authorization can do the opposite: make consent visible, searchable and commercially useful.

Method note: Likeness Index distinguishes marketplace claims from independently verified facts. Availability, prices and restrictions should always be confirmed with the originating source before a license or production decision.