Deepfakes and AI: can existing IP rights provide protection?

While the UK government stalled in its plans to respond to the advances of AI with legislative change in its March 2026 Report and Impact Assessment on Copyright and Artificial Intelligence (the Report), celebrities have turned to trade mark law to combat deepfake imitations of their voices, catchphrases and appearance.

Around the same time that the Report was published, darts player Luke Littler applied to trade mark his face with the UKIPO, while footballer Cole Palmer’s UK registrations of his face, name, nickname and signature goal celebration were granted in November last year. Last month, in what appears to be the broadest effort yet to protect both voice and appearance against AI misuse, Taylor Swift filed three US trade mark applications: an on-stage image of her during the Eras Tour and two short voice clips of her saying “Hey, it’s Taylor Swift” and “Hey, it’s Taylor”.

The Report contemplates introducing a US-style digital replica or personality right into UK law to protect an individual’s likeness. Before committing to either option, or engaging in the detail of how each might work, the government intends to launch a consultation this summer to seek opinions on how best to address the harms posed by digital replicas without unduly restricting innovation.

This development prompts a broader evaluation of whether a new, dedicated right is necessary, and how well trade mark law can serve as an improvised solution to address unauthorised use of identity. Here we consider the position under English law (however, given broad similarities between intellectual property laws of many jurisdictions, much of what we say will have wider relevance).

The legislative gap

Generative AI can replicate a person’s voice and face with unsettling realism. Public figures are particularly vulnerable to imitation, as their extensive digital footprints provide abundant training material for such systems. Taylor Swift has herself been targeted by malicious applications of AI, most notably in 2024 when AI-generated sexually explicit images of her reportedly circulated on X, prompting renewed calls for stronger legal safeguards.

Unlike jurisdictions such as the US, where publicity rights are recognised at state level, or Germany, which has long recognised personality rights, UK law provides no standalone statutory protection for a person’s image or identity.

UK courts have also been reluctant to expand existing doctrines, such as passing off, to fill this gap. In Elvis Presley Enterprises Inc v Sid Shaw Elvisly Yours [1999] RPC 567, Simon Brown LJ in the Court of Appeal cautioned against assuming that only a celebrity “may ever market (or license the marketing of) his own character,” reflecting concern that such recognition would too readily create monopolies over identity.

Copyright law offers limited recourse against deepfakes. To establish infringement, a claimant must show that a substantial part of a protected work has been copied. Generative AI, however, can produce convincing likenesses without copying a single source, instead synthetising new material from its training data. Therefore, even where aspects of AI development are likely to infringe copyright, this framework does little to protect individuals whose likenesses are created independently of any identifiable original works.

Trade mark law as a workaround

Trade mark infringement does not depend on copying, but on whether a trade mark’s ability to distinguish the commercial origin of a product or service has been harmed. Deepfakes are essentially a form of false attribution, presenting synthetic content as authentic. This creates a real risk that audiences will be misled as to the association of celebrities with particular products or services.

As a result, trade marks represent a potential enforcement tool against deepfakes. The modern definition of a trade mark is deliberately broad, encompassing any sign capable of distinguishing goods or services, including sounds, video clips and images. Such breadth explains why celebrities have had success registering catchphrases, gestures and specific visual depictions associated with their brand.

Typically, trade mark infringement turns on whether consumers are confused about the commercial origin of goods or services. This focus becomes especially valuable in endorsement contexts. If a deepfake suggests that a celebrity has approved or is associated with a product, it risks creating a false commercial connection – that is, precisely the type of harm trade mark law is designed to prevent. 

For well-known marks, protection extends further. The law can address not only confusion, but also uses that take unfair advantage of, or cause detriment to, the reputation associated with a mark. This is particularly significant in the deepfake context, where harm may arise less from consumer purchasing decisions and more from reputational damage: for example, where a fabricated image places a celebrity in a degrading or offensive setting.

A registered mark can also be leveraged against social media deepfakes through takedown procedures, bypassing many of the practical difficulties associated with infringement litigation. Platforms typically structure their IP complaint systems around registered rights, and the submission of a registration number can often trigger the swift removal of content without protracted evidential exchanges. This provides a degree of speed and certainty that is crucial given deepfakes can spread rapidly and virally online.

The limitations of trade mark law

However, trade mark law was not designed to protect likeness and has limitations when used for this purpose. A clear obstacle to registration is distinctiveness. A trade mark must function as a badge of origin, identifying goods or services as coming from a particular undertaking. This has long posed difficulties for celebrity marks. In Elvis Presley Enterprises [1999] RPC 567, the Court of Appeal held that the name “Elvis” did not indicate trade origin, but instead signified the singer himself. The Court warned that a celebrity who waits too long may find that the very fame of their persona undermines its distinctiveness.

Swift’s filings illustrate this tension. Her sound marks - “Hey, it’s Taylor” and “Hey, it’s Taylor Swift” - are short, distinctive phrases used in a specific promotional context which arguably function more like traditional slogans. The question is whether consumers would perceive them as indicators of commercial origin, or simply as Taylor Swift speaking.

Her on-stage image mark faces a different challenge. As the General Court’s decision in Jan Smit v EUIPO (T-82/20) demonstrates, consumers may view a naturalistic image as merely depicting a person rather than indicating trade source. The more recognisable such an image becomes, the greater the risk that it is treated as decorative rather than distinctive.

Even where registration is secured, the scope of protection is limited. Trade mark registrations are confined to specified goods and services and most infringement offences require unauthorised use in the same or similar fields. Trade mark law also resists defensive stockpiling: the mark must be put to genuine use in the specified fields within five years or risks being revoked for non-use. As the UK Supreme Court made clear in Sky v SkyKick [2024] UKSC 50, overly broad filings without a genuine intention to use are also vulnerable to challenge on the basis of bad faith.

A further uncertainty lies in whether deepfakes amount to “use in the course of trade.” Trade mark infringement requires a commercial context, but deepfakes often circulate in a blurred space between expression, entertainment and monetised content. Some uses, such as fabricated endorsements, will clearly qualify; others, including viral videos, parody or anonymously shared content, may not. While such material can generate revenue indirectly through advertising or engagement, the link to a specific commercial offering is often difficult to establish. Whether courts will treat this as sufficient “trade use” remains an open question.

Passing off presents a parallel route, but similar challenges arise. In particular, the requirement of misrepresentation (that is, that the public has been deceived into believing there is a commercial connection) can be difficult to satisfy where deepfakes are ambiguous, humorous or consumed sceptically. These evidential difficulties are compounded by the practical challenge of identifying creators, given that such content often circulates pseudonymously across global platforms.

What next?

The UK government’s decision to defer reform leaves a clear doctrinal gap. While consultation is promised for the summer, as yet there are no new statutory protections on the horizon. Into that gap have stepped celebrities and their advisers, repurposing trade mark law as a means of controlling unauthorised use of identity. Such developments raise a fundamental policy question: whether the emergence of protection should be left to judicial evaluation of pre-existing doctrines or directly addressed through a coherent, purpose-built regime? 

Other jurisdictions have begun to confront this question. Denmark, for example, proposes inserting protections against imitations of personal characteristics and performers into its existing copyright framework, enforceable through its Digital Services Act. China has introduced more discrete regulatory rules requiring consent for the creation of AI-generated likenesses and mandating the labelling of synthetic content. While each approach has its limitations – Denmark’s can be criticised for being conceptually awkward within copyright's traditional subject-matter and China's for containing prohibitions which are too vague and broadly defined to offer a replicable model – legislative inaction is unlikely to prove a sustainable alternative. Until UK lawmakers intervene, individuals will remain reliant on the incremental adaptation of ill-fitting doctrines, rather than the protection of a regime designed for the confronting realities of synthetic media.