Who Owns an AI Celebrity? Taylor Swift, Matthew McConaughey, Erling Haaland, and the Limits of Trademark and Publicity Rights

Erling Haaland had a very “online” World Cup. As Norway ”rowed” its way through the tournament, social media reimagined its star striker as a Viking leader, a K-pop idol, and the star of an endless supply of memes, TikTok videos, and AI-generated videos. Most were obvious jokes. Others raised a more difficult question: when anyone can reproduce a famous face or voice, who controls the replica?

Taylor Swift is testing one answer. In April 2026, TAS Rights Management filed trademark applications for sound marks consisting of the spoken phrases “Hey, it’s Taylor” and “Hey, it’s Taylor Swift” for specified entertainment services. Readers can listen to the audio specimen submitted with the “Hey, it’s Taylor Swift” application. Matthew McConaughey has gone further, obtaining registrations for several voice and motion clips associated with his persona, including his familiar “alright, alright, alright” delivery.

These strategies are creative, but neither celebrity has trademarked a voice, face, or personality in the abstract. Instead, they have added another layer to a legal system that still treats identity, expression, and commercial source as distinct concepts.

A Voice Is Not Automatically a Trademark

Trademark law protects source identifiers. See 15 U.S.C. § 1127; Pirone v. MacMillan, Inc., 894 F.2d 579, 581–84 (2d Cir. 1990). Sounds, whether a celebrity’s voice, Homer Simpson’s iconic “D’Oh”, or Darth Vader’s breathing, can identify the source of goods or services just as effectively as a word mark. Yet, a sound may be familiar, or even highly recognizable, and still fail to function as a trademark if consumers do not perceive it as identifying the source of goods or services. See In re Vertex Grp. LLC, 89 U.S.P.Q.2d 1694, 1700–02 (T.T.A.B. 2009). Consumers, for example, might hear a synthetic voice and think “Taylor Swift” without understanding that voice to identify the commercial source of a particular service.

That is why Swift’s applications matter most where recognition becomes branding. If the applications mature into registrations, they could strengthen claims against direct copies or confusingly similar uses of the applied-for phrases for related services. They would not, however, automatically reach every AI-generated song, interview, or joke that merely resembles Swift’s voice. The same limitation applies to McConaughey’s strategy: a registration can help define a legal perimeter, but it does not transform an entire personality into a trademark.

Old Rules, New Replicas

Courts were grappling with synthetic celebrity identity long before generative AI. For example, in Waits v. Frito-Lay, Inc., a Doritos commercial used a singer deliberately selected to imitate Tom Waits’s distinctive voice. 987 F.2d 1093, 1097–99, 1100–03, 1106–11 (9th Cir. 1992), abrogated on other grounds by Lexmark International, Inc. v. Static Control Components, Inc., 572 U.S. 118, 129–34 (2014). The Ninth Circuit upheld liability under California’s voice misappropriation doctrine and recognized a federal false-endorsement claim because listeners could believe Waits had performed in or approved the advertisement.

The technology has changed, but the commercial question remains familiar. In Young v. NeoCortext, Inc., a federal court allowed a California right-of-publicity claim against the developer of the Reface app to proceed. 690 F. Supp. 3d 1091, 1095–96, 1101–04 (C.D. Cal. 2023). The plaintiff alleged that the app used his name and likeness to promote the product and sell subscriptions for a face swapping app. The court did not hold that every face swap is unlawful. Rather, it concluded that the claim could proceed where the celebrity content allegedly helped market the app.

Lehrman v. Lovo, Inc., shows the opposite limit. 790 F. Supp. 3d 348, 355–56, 365–70 (S.D.N.Y. 2025). Two voice actors alleged that an AI company created and sold synthetic clones of their voices. The court dismissed their federal trademark claims because the actors had not adequately alleged that their voices functioned as source-identifying marks. The lesson is not that AI cloning is automatically lawful. Rather, the viability of a claim depends on what was copied, how it was used, and what consumers were likely to understand. The case also highlights the limits of trademark law when a use is not clearly source-identifying.

The Commercial Line Matters

As technology improves, we will continue to see more imitations and recreations of celebrity voices and likenesses. Entire industries have already emerged around that capability. James Earl Jones, for example, may have passed away, but his voice remains synonymous with Darth Vader. Before his death, he granted Lucasfilm the rights to recreate his voice using AI. Likewise, the company ElevenLabs operates an “Iconic Voice Marketplace” where creators can license recognizable voices for projects, including those associated with figures such as Sir Michael Caine, Stan Lee, and John Wayne. Voice recreation is no longer a novelty. It is becoming a standard part of the content ecosystem.

As courts confront AI-generated content, the purpose of the use often matters as much as the technology itself. Is the content parody, commentary, entertainment, or a commercial endorsement? California’s leading transformative-use decision, Comedy III Prods., Inc. v. Gary Saderup, Inc., asks whether a celebrity likeness is merely raw material used to create a new work or instead remains the work’s “sum and substance.” 25 Cal. 4th 387, 406–07 (2001). That framework maps reasonably well onto modern AI content.

• Commentary, satire, and parody generally have stronger speech protections. An obviously fictional Haaland-as-Viking video commenting on World Cup mania is different from a realistic video presented as an authentic interview.
• Commercial use raises greater risk. Using a celebrity replica as the recurring attraction for a paid app, subscription, voice model, or face-swapping template creates a more direct commercial connection and risks misleading consumers.
• False endorsements present the clearest legal concern. An AI-generated Swift, Haaland, or McConaughey recommending a product can implicate both false-endorsement and right of publicity claims, even when the original photograph or recording was not copied.

The fact that a post earns platform revenue does not, by itself, transform the use of a celebrity into advertising. Courts generally focus on whether the celebrity replica is being used to attract customers or sell a particular product, service, or subscription. See Cal. Civ. Code § 3344(e); Comedy III Productions, 25 Cal. 4th at 396–97. But as the use moves closer to selling a product, attracting paid subscribers, or creating the impression of genuine approval, it becomes increasingly difficult to characterize the content as merely discussing or depicting the celebrity.

A Patchwork, Not a Silver Bullet

No single doctrine fills every gap. Copyright protects a particular photograph, video, or sound recording, but not a face, voice, or performance style in the abstract. Trademark law focuses on source identification and the likelihood of consumer confusion. State right-of-publicity laws more directly address commercial exploitation of identity, but their scope and defenses vary.

Legislatures are beginning to respond. Tennessee’s ELVIS Act expressly protects readily identifiable simulations of a person’s voice and can reach technologies primarily designed to generate an unauthorized voice or likeness. At the federal level, the NO FAKES Act of 2026 Act would create a nationwide framework governing certain highly realistic digital replicas while preserving protections for news, commentary, criticism, satire, and parody. The Senate Judiciary Committee reported the bill favorably in June 2026, but it has not yet become law.

Until Congress enacts a more uniform federal framework, we are likely to see more celebrities turn to trademark filings and other existing legal tools to protect their names, images, and likenesses. For now, advertisers, influencers, and content creators must navigate a patchwork of overlapping laws, particularly when their use of AI-generated content has a commercial purpose. That patchwork helps explain why Swift’s and McConaughey’s trademark filings matter without resolving the broader problem. Sound and motion marks may create useful federal claims in the right circumstances, but publicity rights, copyright ownership, contracts, and platform policies all remain important parts of the analysis.

For advertisers, influencers, and content creators, an “AI-generated” label is not a license. Before publishing, it is worth asking: Is the celebrity merely the subject of the content, or the reason the product has value? Could viewers mistake the replica for an authentic performance or endorsement? And is the use tied directly to advertising, subscriptions, or sales? The answers to those questions will often matter far more than the technology used to create the content.

AI may make a celebrity endlessly reproducible. The law still asks a familiar question: are you talking about the person, or using the person to sell something?