The recent debut of “AI actors” – brought into the spotlight by ‘Tilly Norwood’ – has provoked both fascination and unease in equal measure. For some, these digital performers mark the next frontier of creativity: tireless, perfectly consistent and infinitely adaptable. For others, they raise uncomfortable questions about authenticity, artistic integrity and, crucially, legality.
The technology has clearly outpaced the law. As with many creative revolutions, two questions underpin the debate: are we free to create it, and can we protect it following its creation? At present, the first is considerably harder to answer than the second.
Freedom to operate
Anyone seeking to develop an AI actor must first ensure they have the legal “freedom to operate”. That is, confidence that the digital persona has not been created through the unauthorised use of others’ intellectual property. This can be where trouble begins.
The process by which AI models generate visual likenesses or human forms is both technically complex and commercially unclear. The datasets used to train these systems are vast and, in many cases, undisclosed. As a result, determining whether a digital likeness is the product of lawful training or inadvertent infringement can be close to impossible.
The Getty Images v Stability AI case, currently before the courts, illustrates this problem aptly. Getty alleges that Stability AI used millions of its copyrighted photographs without permission to train its image-generation model. Stability AI disputes this, but the case highlights a broader uncertainty: even if a developer acts in good faith, they may have no practical means of proving that the resulting output is clean.
For those seeking to invest in AI-generated performers, that opacity creates material legal risk. The absence of a clear provenance trail means that every digital face carries with it a potential claim, however remote, from a photographer, performer or estate. One can easily imagine those opposed to the rise of AI actors leveraging this uncertainty to challenge their deployment, particularly in high-profile productions or advertising campaigns where reputational risk is a factor.
Until there is greater transparency in training data, or statutory reform to clarify when and how copyrighted material may be used for AI training, this “freedom to operate” question could remain a real deterrent to mainstream adoption.
Protecting the persona
Ironically, protecting an AI actor once created is comparatively straightforward. The law already provides mechanisms to safeguard distinctive visual elements, including motion trade marks, registered designs and copyright.
Trade mark registration offers perhaps the most powerful tool. A distinctive face, animation sequence or logo, associated with a digital actor, can be registered as a trade mark, provided it is distinctive and can be adequately represented using generally available technology. The UK Intellectual Property Office and the EU Intellectual Property Office have both granted such marks in recent years. Provided they are renewed, trade mark registrations can last forever.
Copyright and design law also remain useful, particularly because they provide for automatic protection. In Europe, unregistered design rights can protect the appearance of a digital character for up to three years, while registered designs extend protection further.
This is, in essence, familiar territory. The same frameworks that protect cartoon mascots, brand icons and video game avatars can be adapted to digital humans. The novelty lies not in the law, but in the degree of realism now achievable.
An AI actor’s “image” could, for instance, be protected much like a fictional character such as Mickey Mouse or Lara Croft. Both of these names are registered trade marks, in the name of Disney Enterprises, Inc. and CDE Entertainment Ltd respectively. The difference is that these digital performers may evolve continuously through machine learning, complicating the question of what exactly is being protected at any given time.
Who owns the actor?
Ownership is where the law becomes even more complex. Copyright traditionally vests in the “author”, meaning the human(s) who created the work. The problem for AI-generated content is that there may be no such author.
Under current UK law, the Copyright, Designs and Patents Act 1988 attributes authorship of a computer-generated work to the person “by whom the arrangements necessary for the creation of the work are undertaken”. This was a sensible provision in 1988, when computer-generated art meant little more than digital graphics produced by humans using software tools. Whether that formulation still works in the age of generative AI is open to debate.
Many jurisdictions do not recognise copyright in works without human authorship at all. The United States Copyright Office, for example, has made clear that AI-generated works, without significant human creative input, are not registrable. China, conversely, has shown some willingness to recognise such works under broader interpretations of “authorship”. The global picture, therefore, is fragmented. As far as designs are concerned, the current Consultation on Changes to the UK Designs Framework, has revealed the government’s preferred option to be the removal of exiting provisions to protect computer-generated designs without a human author. The Consultation notes that “most leading AI nations do not provide protection for computer-generated designs without a human author. This includes the US and most EU Member States”. .
For clients, this means one thing: contracts matter more than ever. Ownership of the intellectual property in an AI actor, including its likeness, voice, and performance outputs, must be defined contractually from the outset. If multiple contributors are involved (for example, a developer, an AI model provider and a commissioning brand), rights must be clearly allocated between them.
Without clear agreements, disputes are inevitable. The history of music and film offers countless examples of collaborators falling out over ownership, and that is before introducing the added complexity of a non-human creator.
The personality question
Beyond IP, a further layer of complexity arises in relation to personality and performer rights. Some jurisdictions, including certain EU member states, recognise “personality rights” i.e. the right to control the commercial use of one’s likeness.
If an AI actor is modelled, even loosely, on the appearance or mannerisms of a real individual, there is a potential claim for misappropriation or infringement of image rights. The recent dispute between Scarlett Johansson and OpenAI over the alleged imitation of her voice for the “Sky” persona in ChatGPT underscores how sensitive this area has become.
Developers must therefore tread carefully when designing digital personas, particularly if the resemblance to any real person is more than coincidental. In the absence of a general “right of publicity” in UK law, such disputes would likely be pursued through claims in passing off or data protection, but the reputational damage alone could be considerable.
A long road to the spotlight
The creative industries are rarely short of legal grey areas, but AI actors inhabit a particularly foggy patch. Their potential is evident: they could front advertising campaigns, perform in films, or interact with audiences in real time. Yet the uncertainty surrounding their lawful creation and the absence of clear ownership frameworks is likely to keep them on the margins for now. Pressure from acting unions is also a factor.
The road to mainstream acceptance will require a combination of technological transparency, legal reform and contractual rigour. The law is beginning to move in this direction: the UK government’s ongoing review of copyright and AI aims to clarify how existing frameworks apply to generative systems. But it will take time.
Until then, the greatest barrier to AI actors taking centre stage will not be consumer scepticism or lack of creative ambition. It will be the unresolved question of whether the very process of creating them is lawful.
