On April 24, 2026, Taylor Swift filed a trademark application for the phrase “Hey, it’s Taylor Swift”, specifically, the sound of her spoken voice, alongside an image of herself onstage in her signature sparkly jumpsuit. Matthew McConaughey did something similar months earlier with “Alright, alright, alright.” The headline writes itself: celebrities are fighting back against AI deepfakes using trademark law.
It is the most recent, and possibly most consequential, example of a recurring pattern in IP law: the reflexive instinct to reach for the nearest available exclusionary right whenever a new technological threat appears, without asking whether that right was designed to do the job being demanded of it.
The harm AI voice cloning causes is real. The legal response being fashioned is the wrong one, and if it succeeds, it will embed a permanent doctrinal error into trademark law that will far outlast the AI moment that prompted it.
Trademark attorney Josh Gerben, who flagged Swift’s filings, identified the theory plainly: copyright law protects recorded performances but requires reproduction of an existing recording. AI doesn’t copy an existing recording, it generates a new one that sounds like the celebrity. Hence the gap. Trademark, by contrast, stops not just identical uses but anything “confusingly similar,” and does not require copying at all. Trademark protection also does not expire. Therefore, the appeal.
The logic is superficially elegant but it collapses the moment you ask the foundational question trademark law actually requires you to ask.
Trademark law exists to facilitate a consumer’s reliable identification of the source of goods and services. That is not an incidental feature of the doctrine, it is its entire constitutional and policy justification. The source-identification function is what distinguishes trademark’s exclusionary right from every other form of IP exclusion. It is what makes trademark legitimate.
The relevant inquiry is therefore not: “can this attribute possibly identify a source?” That question, if it were the test, would make almost anything registrable. Justice Breyer in Qualitex v. Jacobson Products made exactly this error, extending the meaning of “symbol or device” to include “anything capable of carrying meaning” without examining whether the extension was consistent with trademark’s purpose. In Two Pesos v. Taco Cabana, the Supreme Court of the United States extended inherent distinctiveness to trade dress, the total image and overall appearance of a business, without asking whether consumers actually relied on those features to identify source, as distinct from simply finding them attractive. As Glynn Lunney argued in The Trade Dress Emperor’s New Clothes, this was a judicial improvisation the legislators never intended and the statutory text never authorised.
The right question is: divorced from the name and logo that already perform the source identification function, does an average consumer, in ordinary commercial life, rely on a celebrity’s voice alone to identify the source of a good or service in trade? The answer is almost never yes. And crucially, even where some secondary meaning has accrued, even where consumers do associate that voice with that person, that association is beside the point when a dedicated source identifier already exists. It is not objectively reasonable to assume that, in the presence of the boldly written name “Reese’s” in its distinctive font, consumers are simultaneously relying on the orange wrapper to identify source. The dedicated identifier makes the secondary attribute redundant as a trademark.
Swift’s voice is not how consumers identify the source of her goods and services in trade. Her name is. Her voice is the person.
What Swift is actually protecting, what she should be protecting, is not a source indicator for goods in commerce. It is her identity: her voice, as an element of the cultural grammar of our time. That is a right of publicity / personality rights claim.
Indian courts have confronted exactly this situation. In Arijit Singh v. Codible Ventures LLP, the Bombay High Court granted an ex parte ad-interim injunction prohibiting AI tools from exploiting the singer’s name, image, voice, and personality traits. The Delhi High Court in Anil Kapoor v. Simply Life India and Jackie Shroff v. The Peppy Store followed the same course. One may criticize those orders, and there is much to criticise, given that they were passed ex parte without the courts weighing the celebrity’s interest in controlling their identity against the freedom of expression and the commercial interests of others. But at least those courts named the right they were protecting. They did not hijack trademark law to do it.
The US filing does precisely that, and the motive is transparent. A right-of-publicity claim under US state law is territorially limited, inconsistently defined, and in many states not perpetual. A federal trademark registration, by contrast, is nationally uniform, broadly enforced, and does not expire so long as the mark is in use. The trademark route is not a more accurate legal response to the harm, but it is a more powerful one.
The perpetuity point deserves emphasis because it is the feature of this development that most alarms. Copyright enters the public domain. The copyright bargain, limited monopoly in exchange for eventual public access, is what makes the IP system constitutionally and morally defensible. Trademark has no such expiration. A successfully registered voice mark would create a permanent exclusionary right over a sonic identity. Combine that with trademark’s “confusingly similar” standard, far broader than copyright’s reproduction requirement, and you have a sweeping, perpetual monopoly over an attribute that is simultaneously part of our shared cultural life.
The rhetoric of natural rights and unjust enrichment obscures what celebrity identity protection is really about: who controls the production and circulation of meaning in our society. Celebrities are not just private individuals. They are part of our shared cultural narrative, the grammar through which we communicate, critique, and make sense of the world. Granting them a permanent trademark over their voice is not protecting an income stream. It is privatising an element of cultural infrastructure.
And there is an uncomfortable secondary question that Michael Madow asked in 1993 and that nobody has answered satisfactorily since: on what basis does identity juxtaposed with fame itself confer deserving? Fame is not produced by individual effort alone, it is partly arbitrary, structurally determined, a product of timing, luck, and the collective whims of media and markets. The legal system is not equipped to distinguish degrees of deserving fame. The trademark system most certainly isn’t.
The answer is legislative one which gives a time-limited, clearly defined right against non-consensual AI voice cloning and likeness replication is the correct response. It would address the harm directly, without distorting trademark law’s purpose, without creating perpetual rights over cultural identity, and without foreclosing legitimate uses of AI-generated content that is neither deceptive nor defamatory.
Until that statute exists, courts will face celebrity trademark applications like Swift’s and be tempted to let them through, after all, the harm is sympathetic, the applicant is famous, and the doctrinal barriers are porous enough to allow passage. That would be a mistake. What trademark law cannot do, what it was never designed to do, is substitute for the IP system’s missing piece.
The answer to a gap in the law is to fill the gap. Not to flood the adjacent terrain.
Image created on Dall-E
