Likeness rights — the legal protection of individuals' right to control how their name, image, voice, and other identifying attributes are used — emerged from the early twentieth century right of publicity doctrine as a mechanism for preventing the commercial exploitation of recognizable identities without consent. For most of their history, likeness rights were primarily the concern of celebrities and public figures who faced commercial appropriation of their fame. The AI era has changed this fundamentally. Technologies capable of generating photorealistic synthetic images, videos, and audio of real people — using only publicly available data as input — have transformed likeness appropriation from a problem of celebrity exploitation into a collective challenge affecting every person with a digital presence, and potentially every person whose face or voice has ever been captured in a photograph or recording.
The core problem is the collapse of the representational barrier between real and synthetic identity. Historically, a photograph was evidence of an event that occurred; a recording captured a performance that was given; a quotation reflected words that were said. These were representations with inherent evidentiary constraints — to falsify them required significant effort, skill, and resources. Generative AI eliminates these constraints. High-quality synthetic images, videos, and audio of real people can be produced by anyone with modest technical skills and access to consumer-grade hardware. The social and legal infrastructure premised on the assumption that representations of people correspond to their actual behavior is now structurally inadequate.
The consequences operate at multiple levels. For individuals, the most immediately harmful application is non-consensual synthetic pornography — the creation of sexually explicit images or videos depicting real people without their consent. This is overwhelmingly used as a tool of harassment and abuse, disproportionately targeting women. The harm is severe: victims report profound violations of dignity, reputational damage, psychological trauma, and the practical impossibility of containing the spread of images once they are published. Surveys of revenge porn and synthetic pornography prevalence consistently find that women are targeted at rates an order of magnitude higher than men. Some jurisdictions have criminalized non-consensual synthetic pornography — the United Kingdom, several US states — but enforcement faces severe challenges given the global distribution of content creation and hosting and the speed with which synthetic content can be generated and spread.
Beyond non-consensual pornography, AI-generated likeness manipulation enables sophisticated disinformation. Deepfake videos depicting public figures — politicians, business leaders, military officers — making false statements pose direct threats to political discourse, financial markets, and international stability. The 2024 US presidential election cycle saw the deployment of AI-generated audio mimicking candidates' voices; similar techniques have been used to produce false video statements attributed to political leaders in multiple countries. The deeper epistemic problem is not any specific deepfake but the general undermining of visual and audio evidence as epistemically reliable — once it becomes widely understood that any image or recording might be synthetic, the evidentiary value of genuine recordings is compromised.
At collective scale, likeness rights in the AI era are inseparable from questions about the governance of AI systems that generate synthetic representations of people. The technical capabilities to generate such representations are concentrated in a small number of AI development companies; the deployment of those capabilities is diffused across millions of applications; the harm falls on billions of potential subjects. This asymmetry between the concentration of capability and the diffusion of harm defines the governance challenge. Existing right of publicity doctrine, developed to address commercial exploitation by identified defendants with traceable relationships to the exploited identity, is structurally inadequate for a problem in which harms are generated by distributed actors using commoditized tools.
Law 4's stewardship obligation in this domain requires governance frameworks capable of operating at several levels simultaneously. At the capability level, AI developers bear responsibility for the foreseeable harms of the systems they build and deploy; this includes the obligation to implement technical safeguards against the most clearly harmful applications of likeness generation (non-consensual pornography, non-consensual impersonation). At the platform level, content distribution systems bear responsibilities for detecting and removing non-consensual synthetic content, including obligations proportionate to the scale of distribution. At the legal level, substantive rights against likeness appropriation and non-consensual synthetic depiction — rights that extend beyond celebrities to all persons — must be backed by enforcement mechanisms capable of reaching across jurisdictions. At the technical level, content authentication systems (cryptographic provenance standards, watermarking, model fingerprinting) can help maintain the distinction between real and synthetic representation, preserving the evidentiary value of genuine recordings.
Law 0 — the foundational law of being — is directly implicated because the capacity to generate convincing synthetic representations of a person enables the effective theft of their identity in the deepest sense: others can speak with their voice, appear in their image, enact with their likeness scenarios that they never chose to enact. This is not merely a reputational harm or a property right violation — it is an attack on the integrity of the person's presence in the world, on the basic correspondence between what a person is and how they appear to others. At collective scale, when this capability is widely available and insufficiently governed, it erodes the foundations of social trust in ways that compound beyond individual harms.
The governance challenge is ultimately about stewardship of a shared epistemic infrastructure. The capacity to trust that representations of people correspond to their actual conduct is a collective good — not owned by anyone, but necessary for everyone. AI-enabled likeness manipulation is a direct threat to that collective good. Protecting it requires treating likeness rights not merely as individual property rights but as a component of the shared infrastructure of social reality — something that governance institutions are obligated to protect on behalf of everyone, not merely on behalf of those with sufficient legal resources to enforce their individual claims.