She has a name, a face, a social-media presence, and reportedly a film role. Talent agencies have circled her. What she does not have is a body, a history, or an existence: Tilly Norwood is an “actress” generated entirely by artificial intelligence. Her arrival has provoked one of the fiercest debates the entertainment world has seen since the actors’ strikes — and, more importantly for anyone who works with intellectual property, it has exposed a hole at the centre of the law. Our rules on identity were built to protect real people. They were never designed for a person who was manufactured.
This briefing looks past the Hollywood drama at the legal frontier Tilly Norwood represents: who owns a synthetic persona, what protects the real performers whose work trains these systems, and why “digital replicas” are about to become a legal problem for every industry — not just film. A note at the outset: this concerns rapidly evolving areas of foreign law and is general commentary only, not legal advice.
What happened
Tilly Norwood was created by Xicoia, an AI “talent studio” spun out of a production company. Presented as a synthetic performer — complete with an online profile and, reportedly, interest from talent agencies and a film role — she was pitched as a new kind of celebrity and a way to cut production costs. The reaction was immediate and hostile. The performers’ union SAG-AFTRA condemned the project, arguing that Tilly “is not an actor” but “a character generated by a computer program that was trained on the work of countless professional performers — without permission or compensation.” Prominent actors voiced alarm; unions in the UK and Canada joined in. The controversy crystallised a fear that had driven the 2023 strikes: that synthetic performers could displace human ones — and do so using the very human performances they were trained on.
Whatever one’s view of the artistic question, the legal questions are genuinely hard.
The problem: our law protects real people, not manufactured ones
The relevant body of law is the right of publicity — often called “name, image and likeness” (NIL) rights — which lets a person control the commercial use of their identity. It is the law that stops a company using your face to sell a product without consent. But it was built around a core assumption: that there is a real, identifiable person whose identity is being used. An AI performer fractures that assumption in two different directions, creating two distinct legal problems.
Problem one — the training data. The first and strongest line of attack is not about the synthetic character at all; it is about how it was built. If an AI performer was trained on the images, faces and performances of real actors without their consent, those actors may have claims — for misappropriation of likeness, violation of publicity rights, and (echoing the AI-copyright cases now moving through the courts) unauthorised use of their work as training data. As one actor put it, if hundreds of real women’s faces were composited to create a synthetic one, where is their consent — and their compensation? This is the same principle emerging across AI generally: liability increasingly turns not on the output, but on the provenance of the inputs.
Problem two — the synthetic persona itself. Here the law is genuinely uncertain. Publicity rights protect identifiable individuals. A wholly fabricated persona that resembles no single real person may fall through the gap — there is no identifiable plaintiff whose likeness was taken. That is precisely what makes a composite synthetic performer legally slippery: it may be built from many real people while legally resembling none of them closely enough to sue. The law has no settled answer to the question “who owns a face that belongs to no one?” — and that uncertainty is exactly what makes these ventures both attractive to their creators and alarming to performers.
A patchwork of protection — and a race to legislate
Because there is no single, uniform right of publicity — in the United States it is a matter of state law, and internationally it varies enormously — the protection available is a patchwork, and legislators are scrambling to catch up with the technology.
Several developments are worth knowing. Tennessee’s ELVIS Act (Ensuring Likeness, Voice and Image Security) specifically extends protection to a person’s voice and likeness against AI cloning. California has enacted laws (AB 1836 and AB 2602) protecting performers’ digital replicas — covering both living and deceased performers and requiring consent. At the federal level, the proposed NO FAKES Act would, if passed, create a federal property right over a person’s voice and likeness against unauthorised AI-generated replicas — a significant potential shift from the current state-by-state approach. Alongside legislation, SAG-AFTRA’s collective bargaining agreements now require advance notice and negotiation before “synthetic performers” are used in a production, exposing non-compliant producers to grievances. And consumer-protection regulators may have a role: advertising and endorsement rules can be triggered where the public is misled into believing it is dealing with a human.
Internationally, approaches diverge. Some systems, like France, do not recognise a standalone “right of publicity” but protect the right to one’s image and performers’ neighbouring rights through other doctrines. There is, as yet, no coherent global framework — which matters enormously, because AI performers and content cross borders instantly.
The lessons — and they reach far beyond Hollywood
It would be a mistake to see this as a niche entertainment issue. The “digital replica” problem is coming for every sector — synthetic brand ambassadors, AI-cloned executive voices, virtual influencers, automated customer-facing avatars. The legal principles the Tilly Norwood debate is surfacing will apply broadly.
For individuals — your identity is now a replicable asset. Your face and voice can be captured, cloned and synthesised. Protecting them is becoming an active exercise: understanding what rights you hold, watching the fast-moving legislation, and — critically — reading what you sign. Performers and public figures are increasingly asked to license “digital likeness” rights in contracts; those clauses can quietly authorise exactly the AI replication a person would later want to prevent. The consent buried in an agreement is the whole game.
For businesses using AI personas — the exposure is real. A synthetic spokesperson or avatar that is trained on real people, or that too closely mirrors an identifiable individual, invites claims for misappropriation, publicity-rights violations and false endorsement. The safest course is rigorous attention to training-data provenance (did you have the right to use what the model learned from?) and to disclosure (are consumers misled into thinking the persona is human?). As with AI generally, the question is shifting from “what did you make?” to “what did you make it from, and did you have the right?”
For the industries that employ creative talent — the contracts must evolve now. The gap the law is racing to fill will, in the meantime, be filled by contract. Who owns a persona, what a performer has consented to, how a digital replica may be used and for how long — these must be negotiated and documented expressly, because the default legal position is uncertain and shifting.
A concluding observation
Tilly Norwood is a marketing creation, and the temptation is to dismiss the storm around her as a passing controversy. That would miss the point. She is a stress test for a fundamental legal concept — the ownership of human identity — at the moment technology made that identity synthesisable. The law of publicity assumed a real person behind every protected persona; AI has severed that link, and neither the courts nor the legislatures have fully answered what follows. In the interim, the exposure is genuine on every side: for the performers whose work trains these systems, for the companies that build and deploy them, and for the industries drafting the contracts that will govern them. What is certain is the direction of travel. In the AI age, a face, a voice and a performance are no longer purely personal attributes — they are assets, capable of being copied, and the fight over who owns them, and who consented to their use, is only beginning. The businesses and individuals who treat their identity — and their training data — as the valuable, protectable, consent-governed assets they have become will be the ones prepared for what comes next.
Muzy & Meraris LLP publishes this briefing for general awareness. It concerns rapidly evolving areas of intellectual property, publicity and technology law across multiple jurisdictions; it is general commentary only, does not constitute legal advice, and no lawyer-client relationship is created by it. The law in this area is developing quickly and varies significantly by jurisdiction; specific advice should be taken on any particular situation.