Why Copyright Cannot Answer the Whole Question
The question in this chapter is why a realistic AI replica can be harmful even when no one owns the particular pixels or sound in the output. The answer lies in the way U.S. law divides the problem among state privacy and publicity rights, copyright, consumer protection, trademark law, and communications regulation. Mapping those routes helps readers judge a replica scenario by its facts—who is depicted, what was used, where it was released, and what the audience was led to believe—rather than by the label “AI.”
The central difficulty is that a face or voice can carry both personal and economic meaning. A person may want protection from humiliation even if their identity has no commercial value. A performer may want to license a voice while preventing a false endorsement. A copyright owner may want to stop a recording from being copied, although the person heard in it does not own the recording. Current law recognizes pieces of these interests, but it does not place them in one uniform box.
State Law Protects Identity Unevenly
The most direct existing protections are found in state privacy and publicity law. Privacy law generally addresses personal interests such as autonomy, dignity, and freedom from certain forms of intrusion or false portrayal. Publicity law addresses the use of a person’s identity for another party’s advantage, especially when the use exploits commercial value. The categories overlap, but they are not interchangeable everywhere, and the difference matters when a replica causes personal harm without generating an advertisement or sale.
A false portrayal can injure someone by making them appear to have said or done something that never happened. A synthetic intimate image, for example, may attack dignity and reputation even when the person depicted has no endorsement market to protect. An appropriation theory may seem well suited to a replica because it focuses on the unauthorized use of an identifiable aspect of a person. Yet the Office’s survey notes that some jurisdictions limit this kind of claim to commercial purposes or require a showing that the identity has value. The same image can therefore present a stronger claim in one state than in another, even when the emotional injury is similar.
Publicity rights vary in other ways. Some state rules protect a voice, gestures, mannerisms, or a broad set of attributes that identify a person. Others are narrower. Some protect only celebrities, professional performers, people whose identities have commercial value, or particular classes of living or deceased individuals. A law that covers an actor’s commercial persona may not offer the same protection to a student whose face is inserted into an explicit image. A law that protects advertising may not clearly reach a noncommercial post made to harass someone.
Postmortem rights show how deeply these choices differ. The Copyright Office reported that, at the time of its 2024 study, 27 states provided some form of postmortem publicity right. The terms ranged from 20 years to 100 years, and in one state could continue indefinitely while the identity was continuously exploited. Other states provide no equivalent protection or treat the interest as personal to the living individual. This is not merely a technical difference for estates. It determines whether an heir, licensee, or successor can authorize a synthetic performance or challenge a false one after the person’s death.
State law also supplies different answers about speech and intermediaries. Statutes may exempt news reporting, political campaigns, sports broadcasts, satire, criticism, or expressive works. Some protect advertising media that lack knowledge of an unauthorized use; courts in other settings have treated a service as a mere conduit rather than a responsible appropriator. Remedies and jurisdictional rules vary as well. A person’s ability to obtain an injunction or damages can depend on where they live, where the use occurred, and whether the law treats the conduct as commercial.
The result is a patchwork rather than a single national rule. The patchwork is not proof that state law is useless. It contains decades of legal development and can address interests that copyright does not. But it makes protection dependent on geography and legal classification at precisely the moment when a replica can cross state lines and circulate globally within minutes.
The hardest part is often the interface between categories. A person can be harmed in a personal capacity while the same replica is distributed through a commercial service. A post that looks like an individual user’s joke may be tied to advertising or a paid campaign. A synthetic performance may be an expressive work for one purpose and a misleading substitute for a performer for another. The fact that a use appears in a song, video, or social post does not by itself answer whether the person’s identity was appropriated, whether the audience was confused, or whether the use is protected expression.
Geography creates a second interface problem. The person depicted may live in one state, the uploader in another, the platform in a third, and the audience everywhere. A state rule may focus on the person’s domicile, the place of publication, the defendant’s conduct, or the commercial effect. A person who sees a replica spread nationally may still have to locate the legal theory that fits a particular jurisdiction. The internet makes the output borderless, but the available claim may remain local.
This also explains why celebrity status can act as a hidden legal gate. In commercial settings, recognition may help prove that viewers were likely to think a product was endorsed. In a privacy case, however, the injury may not depend on anyone knowing the person’s market value. If a law requires commercial value or consumer awareness, the rule may protect a famous singer from a fake endorsement while offering less certainty to an unknown person targeted by an intimate image or family-voice scam. The visibility of the harm and the legal strength of the claim can therefore move in opposite directions.
Remedies follow the same split. An injunction may be crucial when an image or voice continues to circulate, while a damages theory may depend on proving commercial loss. A platform may be able to remove a post quickly even when the original creator is anonymous, but state law may not clearly tell the platform whether it is responsible for hosting the material. The person’s practical control thus depends not only on whether a legal right exists, but also on who can be identified, what the jurisdiction recognizes, and whether the available remedy matches the injury.
The main routes can be seen as different gates around the same synthetic output:
The diagram’s point is not that a claimant must choose only one gate. A single fake advertisement might implicate a person’s publicity right, a copyrighted photograph, and a false-endorsement theory. The point is that each gate asks a different question. A strong claim through one route does not automatically open the others.
Federal Law Covers Fragments, Not Identity Itself
Copyright is the clearest example of a near match that is not a complete answer. It protects original works of authorship, including photographs and sound or video recordings. If someone copies a protected recording, alters it, or incorporates it into a larger work, the copyright owner may have rights in that work. But copyright does not protect a person’s identity as such. The Copyright Office’s report states that a person’s name or likeness is not a work of authorship, and that a voice by itself is not copyrightable merely because it has been captured in a recording.
That distinction changes the analysis of an AI voice. Suppose a system produces a new recording that sounds like a singer but does not reproduce the actual sounds fixed in an earlier song. The new file may not infringe the copyright in the old recording simply because it imitates the performer. The singer may still have a different kind of claim based on the unauthorized use of their voice, but that claim concerns a personal characteristic, not ownership of the old sound recording. Copyright can therefore protect the container or source material without giving the person complete control over the identity expressed through it.
The Federal Trade Commission Act reaches a different problem: unfair or deceptive acts or practices in commerce. The Copyright Office reports that a digital replica may fall within this framework when it deceives consumers, exploits a creator’s reputation, diminishes the value of existing or future work, reveals private information, or causes substantial consumer injury. This can be useful when a synthetic celebrity advertisement persuades people to buy a product or when a business uses a person’s apparent endorsement without permission.
But the commercial setting is an important boundary. A private replica made to humiliate someone may cause severe harm without being an unfair commercial practice in the relevant sense. The FTC route may also focus on injury to consumers or competition rather than giving every individual a direct, general power to control their identity. It is a protection against certain conduct in commerce, not a universal right of personal likeness.
The Lanham Act, the federal trademark law, can sometimes address false endorsement or another misleading use in commerce. Its central concern is confusion about affiliation, connection, sponsorship, approval, or the origin of goods and services. A fake voice in an advertisement may suggest that a performer approves of the product. A synthetic image in a commercial campaign may create the same impression. Yet a claimant generally must connect the use to commerce and show the kind of consumer confusion the statute addresses. That can be difficult for people who are not famous enough for an audience to recognize the supposed endorsement, and it does little for a replica used outside an endorsement or sales context.
The communications framework is narrower still. The Federal Communications Commission has acted against AI-generated voices in certain robocall scams under its authority over telephony. That matters because a cloned voice can be used to misinform voters, impersonate a public figure, or pressure a family member into sending money. But a rule aimed at robocalls does not automatically reach a synthetic video on a social platform, an AI-generated song, or an intimate image posted on a website. The remedy follows the agency’s jurisdiction and the communication channel, not the identity problem wherever it appears.
Consider four uses of the same person’s cloned voice. In the first, an artist’s existing recording is copied without permission; copyright in the recording may be relevant. In the second, a company uses a newly generated sound-alike in an advertisement; publicity, false endorsement, or consumer-protection theories may matter. In the third, a scammer uses the voice in a robocall; communications law may supply a targeted response. In the fourth, a private user posts the voice in a noncommercial harassment campaign; the strongest route may depend on a state privacy or publicity rule, and the coverage may be uncertain. The technology is similar, but the legal predicates are not.
This is why saying “copyright does not protect identity” should not be turned into the opposite error that copyright is irrelevant. The source photograph, recording, or audiovisual work can still be protected. A person may own that work, license it, or use it as evidence of what was copied. Copyright can also matter when a synthetic output reproduces protectable expression from a particular work rather than merely imitating a person. The narrower point is that copyright ownership of the media does not settle the separate question of who may make the person appear to speak, sing, endorse, or participate in a new event.
Nor does the existence of several partial remedies mean that every harmful replica has no legal answer. The Office’s analysis is a coverage assessment, not a ruling on every possible claim. Facts, ownership, jurisdiction, and later case law matter. The evidence supports a more careful conclusion: current federal law can address particular forms of copying, deception, confusion, and robocall abuse, while state law can address some personal and commercial identity harms, but no single existing framework reliably covers the full range of realistic replicas.
That gap explains the push for federal legislation. A national rule could reduce the importance of where a person lives and could address noncommercial harms that commercial doctrines overlook. But writing the rule would require choices current law often leaves to separate doctrines: how realistic a replica must be, whether ordinary people are protected, whether creation or only public release is actionable, what a distributor must know, how a license ends, and how protected speech is preserved.
The next step is therefore not to add every possible legal label to the list. It is to ask what a federal right would actually own, who would hold it, and where control should attach as a replica moves from private creation into public life.