The Face Is Not the File
This series examines who can control an AI-generated image or voice when the technology can reproduce a person faster than the law can classify the result. The immediate focus is the difference between a person’s identity and the photograph, recording, contract, or model output used to depict it. Understanding that difference gives readers a way to judge the current U.S. legal patchwork, proposed federal protections, platform takedowns, and the trade-off between personal control and protected speech.
The question sounds like a property question: if an artificial system generates a convincing face or voice, who owns the result? But it contains several different questions. Who owns the original photograph or recording? Who is recognizable in the synthetic output? Did that person authorize the use? Was the result kept private or released to the public? Did it imitate a real individual closely enough to mislead people, or merely evoke a style or fictional character? The answers can point to different legal doctrines.
A Replica Is an Identity Claim, Not Just an Edit
The U.S. Copyright Office uses “digital replica” for a video, image, or audio recording that has been digitally created or manipulated to realistically but falsely depict an individual. The definition does not depend on artificial intelligence. A conventional digital technique can make a replica, and an AI system can make an authorized one. “Deepfake” is often used for deceptive or harmful material, but the broader category includes a licensed voice, an accessibility tool, or a creative performance that the person approved.
That distinction matters because realism changes what the output appears to claim. An ordinary drawing may represent a public figure without pretending to be an authentic recording. A digital replica can make an audience believe that a particular person spoke, sang, appeared in an advertisement, or participated in an event. The central legal concern is not simply that pixels or sound waves were copied. It is that a person’s identity has been made to perform a new act.
The relationship can be pictured as a chain:
The source work and the identity are connected, but they are not the same object. Copyright may protect a photograph, an audio recording, or an audiovisual work. It does not automatically turn the person shown or heard into property owned by the copyright holder. Conversely, a person may have an interest in an unauthorized imitation even when they do not own the photograph or recording from which it was made.
This is why an AI-generated face or voice cannot be analyzed by asking only whether a file was copied. A system might generate a new recording rather than reproduce the actual sounds in an old one. It might use a photograph that belongs to someone else but depict a private individual. Or it might create a synthetic advertisement that never copies a particular image while still making viewers believe that a person endorsed a product. Each scenario involves a different relationship between media, identity, authorization, and audience belief.
The Same Capability Can Help or Harm
Digital replication is not inherently abusive. The Copyright Office describes uses that can expand access or preserve creative work. A person with a disability may use an AI-generated model of their earlier voice. A performer may license a voice or likeness for a project and receive compensation. A deceased or non-touring artist may appear in an authorized performance. A filmmaker may use digital techniques to make an actor resemble a historical figure. These uses still raise questions about consent and contract, but the person’s control can be part of the design rather than its casualty.
The harms appear when that control is bypassed or when an audience is misled. Performers and other creative workers worry that replicas can replace paid labor, including background actors and voice actors. That is not only a question of ownership after the fact; it is also a question of who has the power to decide whether human work is substituted and on what terms.
Other harms are more personal. The Office describes sexually explicit or humiliating imagery made without consent, including cases involving people who have no public career and no commercially valuable persona. A private person can be falsely placed in an intimate scene, bullied with a synthetic image, or represented as participating in conduct that never happened. The injury may be dignity, safety, reputation, or emotional well-being rather than lost licensing revenue.
Fraud exploits the same realism in a different way. A cloned voice can appear to come from a loved one asking for money, or from an executive directing a financial transfer. A fake image or voice can also make a celebrity seem to endorse a product. Here the harm spreads across several relationships: the individual is impersonated, the audience is deceived, and a business or victim may suffer a financial loss.
Political and civic uses raise a still wider concern. A synthetic voice can appear to discourage voting, while a fabricated video can make a candidate or public official seem to support a position they never expressed. When authentic recordings become difficult to distinguish from fabricated ones, the damage is not limited to the person depicted. It also affects the public’s ability to decide what evidence to trust.
These examples explain why fame is a poor boundary for protection. A famous singer may face an unauthorized song that attracts millions of listeners, but an ordinary person may face a cloned family voice in a ransom scam. The former is more visible; the latter is not less real. A legal system that protects only commercial value would miss harms that are personal, deceptive, or abusive precisely because they do not depend on celebrity status.
Scale changes the timing of the injury. A photograph or recording may sit unnoticed in an archive until someone chooses to publish it. A synthetic replica can be generated, posted, copied, and recommended across services in the time it takes a person to discover the first version. Removal from one account may not remove the copies already downloaded or reshared elsewhere. That persistence is one reason the legal debate focuses so heavily on distribution and platform response rather than on the technical act of generation alone.
The audience’s belief also matters. A replica can harm a person even when no viewer thinks the underlying file was copied, because the viewer is reacting to the apparent act: the apparent endorsement, confession, performance, threat, or intimate scene. In that sense, the output can function like a false statement made in someone else’s voice. Yet the same visual or audio technique may be used transparently in a documentary, parody, or authorized performance. The important question is not only whether the result looks real, but what the use asks the audience to believe and whether the person agreed to that use.
Authorization is therefore more than a yes-or-no label attached to a model. A person might approve a voice for a particular game but not for political advertising, or agree to a digital double in one film but not to a perpetual library of new performances. A contract may define the setting, duration, compensation, and ability to withdraw or renew permission. These distinctions are why the ownership question eventually becomes a question about scope: who may use which aspect of identity, for what purpose, for how long, and with what safeguards?
The technology can also produce uncertainty without any deliberate impersonation. A generated face or voice may resemble someone by accident, or a user may not know that a clip was synthetic or unauthorized. A rule designed to prevent obvious abuse must still decide how much knowledge a distributor should have before liability attaches. The early distinction between person, identity, source media, and audience belief makes those later choices visible instead of treating every output as the same kind of event.
The Ownership Question Begins With a Distinction
The existing U.S. framework treats these interests unevenly. State privacy and publicity laws may protect dignity, identity, or commercial appropriation, but their coverage varies by state. Some rules focus on advertising or other commercial uses. Some require fame or commercial value. Some recognize postmortem rights, while others provide little or no comparable protection. Speech defenses and intermediary rules also differ. The result is not one national answer to who controls a face or voice, but a set of possible claims whose availability depends on the person, the use, and the jurisdiction.
Federal law covers other fragments. Copyright can reach unauthorized use of a copyrighted photograph or recording, but identity itself is not automatically a copyrighted work. The Federal Trade Commission Act can address unfair or deceptive commercial conduct. The Lanham Act may help with false endorsement when a use in commerce is likely to confuse consumers. Communications law can address certain AI voice-cloning robocalls. Those routes may be powerful in the facts they fit, but none is a general right to control every realistic replica.
That is the problem this series will keep in view. The issue is not whether technology has created a magical new kind of property, nor whether every synthetic resemblance should be forbidden. It is how control should move through a chain of decisions: whether the output is realistically identifiable, whether the person authorized it, whether someone released it publicly, whether a platform can remove it, whether the use is expressive or deceptive, and what remedy is available when harm has already occurred.
The Copyright Office’s 2024 report recommends a new federal right aimed at unauthorized replicas that are difficult to distinguish from authentic depictions, with protection for all individuals. H.R. 8323, the introduced SOUL Act of 2026, offers a different and broader statutory design. Neither document changes the basic distinction between identity and source media. They disagree about how broadly to define the protected interest, how long control should last, how consent should work, how platforms should respond, and how state law should coexist with a federal rule.
So the practical answer is already more precise than “the AI owns it” or “the person owns everything.” Today, a person’s face or voice is not automatically controlled through copyright in the file that captures it. Protection may instead come from privacy or publicity law, a contract, a consumer-protection theory, a communications rule, or the facts of a particular use. Proposed federal laws would make that control more direct, but they would also have to decide where personal authority ends and protected expression begins.
That distinction is the starting point for the next question: if copyright cannot answer the whole problem, what can the existing legal routes reach—and where do they leave the person exposed?