Deepfakes and digital replicas: the UK's legal patchwork

Answer

The UK government's long-awaited report on copyright and artificial intelligence, published in March, was the culmination of a consultation period that had drawn vocal criticism from across the creative industries. Performers' unions, individual artists and prominent public figures had made clear their concerns around the use of their likenesses, voices and personas in training AI systems without consent, and called on the government to enact stronger protections to protect their images and work.

The report, when it arrived, offered a measured response to those concerns, but without providing a wholesale remedy.

While the government ultimately stepped back from its earlier proposal to grant AI developers an unchecked right to train models on copyright-protected works, it left open a broader and arguably thornier question: what protections do individuals have against the misuse of their likeness by AI systems through so-called ‘digital replicas’ or 'deepfakes'?

The report acknowledged both the accelerating sophistication of AI in mimicking real people's faces, voices and mannerisms, and the fragmented state of UK law in addressing such misuse. Calls for the introduction of standalone 'image rights' or 'personality rights', proprietary rights in one's own identity and likeness, have grown correspondingly louder, and the government has committed to further consultation on deepfake harms and formal image rights, expected to launch in summer 2026.

For now, that means there are no immediate plans to legislate. That isn’t inherently surprising. The UK government has often stopped short of direct intervention, expecting the courts to be sufficiently plastic to find creative solutions within existing legal frameworks. The development of passing off, from a narrow 19th Century principle prohibiting the selling of one person’s goods as those of another, to a tool now capable of remedying false celebrity endorsements, is one historic example of that plasticity.

What's certain is that as generative AI continues to proliferate, the stakes of legal reform, whether through legislation or the courts, will only grow.

In the immediate present, however, there remains a pressing practical question. In the absence of statutory image rights in the UK, what rights do individuals currently have to protect themselves against rampant misuse of their image by AI and are those rights doing enough?

What are image rights?

‘Image rights’, otherwise known as ‘personality rights’, give individuals legal control over the commercial use of their name, likeness, voice and public identity. In theory, they reflect the idea that a person's image has commercial value that others shouldn’t exploit without consent.

Such rights already exist in varying forms around the world. In the US, most states recognise legal rights protecting an individual's name, likeness, voice and other personal attributes, often including non-celebrity individuals. Many states have also developed legislation in specific response to AI. In 2024, California enacted Assembly Bills 2602 and 1836 to combat unauthorised digital replicas of performers, and Tennessee's ELVIS Act amended the state's personal rights legislation to protect artists from recognisable AI voice simulations. Germany has comparably strong personality protections through the long-established right to one's own image, which generally prevents the publication or commercial use of an individual's likeness without consent.

By contrast, there’s no codified, standalone image right under English law. Instead, individuals must navigate a patchwork of overlapping legal mechanisms drawn from intellectual property, contract, data protection and the common law, each offering partial protection not designed to deal with the distinctly novel challenges posed by generative AI.

Existing UK protections

Passing off

The tort of passing off is perhaps the most directly relevant tool available under UK law when AI is used to create the false impression that a well-known individual has endorsed or is associated with a product or service. To succeed, a claimant must demonstrate that they enjoy a commercial reputation or goodwill in the UK, that the defendant has made a misrepresentation likely to deceive the public into thinking the claimant is involved, and that real harm has resulted or is likely to follow.

The Court of Appeal's judgment in Rihanna v Topshop [2015] remains the leading authority. Topshop had sold a t-shirt bearing Rihanna's image without her consent. Given her widespread recognition as a style icon who’d previously featured in Topshop promotions, a point of acute practical importance to the case, the court held that a significant section of the purchasing public would assume the garment carried her endorsement. While it ruled in favour of Rihanna, the court ultimately held that English law recognises no general image right allowing a celebrity to control the use of their name or likeness, and that only in specific circumstances, passing off may fill the gap.

The implied endorsement, and not the simple use of Rihanna's likeness, was the actionable wrong. Irvine v Talksport [2002], in which Formula One driver Eddie Irvine successfully sued a radio station for digitally altering a photograph to suggest he endorsed their platform, established the same principle: the false impression of association is what the law catches, not the mere unauthorised use of an image.

A deepfake used commercially in a way that implies a celebrity's endorsement of a product may well satisfy the misrepresentation threshold. What’s certain is that, as Rihanna v Topshop illustrates, any such finding will be highly fact specific. In its current formulation, passing off will also do little for private individuals, who are generally unable to establish commercial goodwill in their name or likeness.

It is, however, worth noting how significantly the tort has been developed since it was first articulated in Perry v Truefit (1842). Back then, the courts established the simple rule that a trader shouldn’t sell their own goods under the pretence that they are those of another. The development of that relatively narrow principle to one now capable of addressing implicit celebrity endorsements is demonstrative of the court’s ability, and willingness, to adapt the law in response to changing times. It may do so again as AI continues to challenge the state of play.

Copyright and performers' rights

Copyright may offer some protection, but its application to AI-generated likenesses is limited and often indirect. There’s no copyright in a person's voice or appearance by default. Where copyright protection arises, it attaches to specific underlying works, such as audio and visual recordings, written prose and photographs, rather than to persons themselves.

Performers' rights, which exist in addition to copyright under the Copyright, Designs and Patents Act 1988 (CDPA), give performers certain controls over how their performances are reproduced and exploited, which may be relevant where AI has trained on audiovisual recordings of a performer without consent, but these are again tied to specific works rather than the individuals more generally.

Developing case law may in time shed light on how existing CDPA concepts apply to AI-generated outputs. The landmark Getty Images v Stability AI, currently before the Court of Appeal, raises important questions about the treatment of protected works in the training of AI models, although its focus isn’t on the unauthorised use of individual likenesses directly. That question remains, for now, unresolved, and may in time be one that requires a legislative response. It wouldn’t be without precedent; Denmark has already taken steps to extend copyright protection to human faces as a means of combating deepfakes, a development worth monitoring as the UK considers its own position.

Trademarks

Trademark registration offers a more targeted form of protection for the distinctive elements of a celebrity's commercial identity, including names, stylised signatures, logos and, in some cases, distinctive visual features. In recent months this route has attracted considerable attention. Luke Littler and Jeremy Clarkson have both registered images of their faces as trademarks at the UK Intellectual Property Office, a response to growing concerns about deepfakes being used to fabricate product endorsements they’d never given. Taylor Swift and Matthew McConaughey have taken similar steps on the other side of the Atlantic to protect their voices. We’ll undoubtedly see more celebrities following suit.

For anything to qualify as a registrable mark, it must function as a badge of origin, differentiating the goods or services of one commercial entity from those of another. Marks must be registered within specific classes of goods and services and are protectable within those (or similar) classes under sections 10(1) and 10(2) of the Trade Marks Act 1994. Where a registered mark has acquired a reputation in the UK, protection may also extend beyond those registered classes under section 10(3), where use by a third party takes unfair advantage of, or is detrimental to, the mark's distinctive character or reputation.

Trademark registrations are a relatively straightforward and proactive step for celebrities to take as a visible signal to prospective infringers of an intention to protect their rights. However, their effectiveness against unauthorised use in the context of deepfakes remains wholly untested.

A further practical constraint is the requirement of genuine use. Trademarks must be used as badges of commercial origin in order to be enforceable and protected from invalidity. For those seeking to register a face or likeness, demonstrating use in that commercially meaningful sense may prove difficult in practice, and failure to do so risks leaving the mark unenforceable when it matters most. In any event, this route remains of little practical relevance to private individuals, who are unlikely to satisfy the badge of origin requirement in the first place.

Contractual protections

Performers, athletes and public figures increasingly manage the commercial use of their image through carefully drafted endorsement agreements and image rights licences that now routinely include AI-specific provisions, such as prohibitions on using recordings or footage to train AI models, requirements for written consent before any digital likenesses are produced and warranties covering the use of AI-generated content in marketing materials.

Collective bargaining has also seen increased calls for the adoption of contractual protections against AI, particularly in the entertainment industry. In the US, SAG-AFTRA's 2023 deal with major studios secured consent and compensation requirements for AI-generated digital replicas of performers. In the UK, Equity, the performing arts and entertainment union, held an indicative ballot in late 2025 in which 99.6% of respondents indicated a willingness to refuse digital scanning without guaranteed contractual protections, marking a significant collective pushback against AI likeness capture.

By definition, however, contracts bind only those who are party to them. While they can offer broader utility (eg, collective agreements may include fighting funds to finance legal action and provisions setting out how members should respond to unauthorised use by third parties), they ultimately offer no direct remedy against those who create deepfakes outside any pre-existing contractual relationship with the individual concerned.

Other mechanisms

Data protection and privacy

UK GDPR may provide additional tools where AI systems have processed personal data, such as images or recordings of an identifiable individual, without a lawful basis. Provided the image is identifiably that of the data subject, affected individuals may be able to challenge that processing as unlawful, object to it or seek its erasure.

Privacy-based claims under the tort of misuse of private information may also lie where AI-generated content reveals or implies private information about an individual. Notably, this cause of action isn’t confined to those with public profiles, attaching to anyone with a reasonable expectation of privacy in the relevant information, making it one of the limited accessible avenues for ordinary individuals who find themselves the subject of deepfake content.

Advertising regulations

The Committees of Advertising Practice and the UK Code of Broadcast Advertising  Codes, rulebooks of the Advertising Standard Authority (ASA) governing non-broadcast and broadcast advertising, prohibit advertising that falsely implies individual endorsement, offering a non-litigious route to challenge commercially deployed deepfakes. However, the ASA can’t award damages and generally operates on a case-by-case basis, offering limited recourse against AI-generated content that can be produced rapidly and at scale.

Criminal offences (online safety acts)

The government has legislated in response to one particularly harmful category. The Online Safety Act 2023 introduced criminal offences, inserted into the Sexual Offences Act 2003, for sharing or threatening to share non-consensual intimate images, including deepfakes; the Data (Use and Access) Act 2025 extended criminalisation to their creation. These measures address only a narrow subset of a much wider problem, but they do demonstrate that the government is prepared to intervene when it deems it necessary.

Defamation and malicious falsehood

Where AI-generated content falsely portrays an individual in a damaging or misleading way, a claim in defamation may be available. A claimant must show that the content is defamatory, that it identifies them and that it’s been published to third parties. The tort of malicious falsehood offers an alternative route where false statements are made dishonestly and cause financial loss.

In practice, the challenge of identifying and pursuing anonymous or overseas creators substantially limits the utility of each of these mechanics, even where the relevant legal thresholds are met.

An uncertain landscape

The overall picture is one of partial protection, available through a creative layering of overlapping legal tools that were never developed with AI-generated content in mind. Trademarks combined with passing off, contractual rights, regulatory enforcement and data protection objections may provide coverage in some circumstances, but they lack the certainty or comprehensiveness that standalone image rights would provide. For now, individuals must continue to assert existing legal rights in response to an environment they were never designed to address, an imperfect answer to a world where technology continues to outpace the law.

We hope to return to this topic as the law in this area continues to develop.

Charlie Osborne is a trainee solicitor at RPC.