Can you trademark your voice and likeness to protect against AI deepfakes?

The Short Answer: Yes, But a Trademark Is Not the Same as a "Right"

You can file U.S. trademark applications covering your voice, your name, your image, and other elements of personal identity with the United States Patent and Trademark Office (USPTO). The applications are real, they are being filed by high-profile figures, and the USPTO has, in fact, registered some of them on the Principal Register. Taylor Swift's filings made headlines in mid-2025 because her legal team used unusually broad specimen language ("audio recordings of musical, vocal, and other entertainment content") and sought registration in two Nice Classes: International Class 9 (downloadable audio and video recordings, downloadable ring tones) and International Class 41 (entertainment services, production of sound recordings). That second application drew formal USPTO review because Section 2(b) of the Lanham Act prohibits registration of a "name, portrait, or signature" identifying a particular living individual without that person's written consent — a quirk that does not automatically apply to voice alone.

Also worth reading: What is the definitive difference between trademark and copyright protection for AI deepfakes, and which legal tool offers better defense for creators? · What counts as genuine use evidence for an EU trademark and how do you protect your brand from cancellation? · What is the EUIPO proof of use checklist for defending an EU trademark against cancellation?

The crucial distinction, however, is that a trademark is a commercial-source identifier, not an absolute property right in your voice. A registered mark gives you the right to stop others from using your voice (or a confusingly similar one) in commerce in a way that would confuse consumers about who endorsed a product or service. It does not, by itself, let you sue a non-commercial deepfake, a parody, news commentary, or a foreign-hosted AI training pipeline. For that broader protection, the operative legal theories are state right of publicity statutes, defamation, Section 43(a) of the Lanham Act for false endorsement, and the still-evolving federal NO FAKES Act.

How a Voice or Likeness Trademark Actually Works

A federal trademark application under Section 1(a) or 1(b) of the Lanham Act must clear several hurdles before the USPTO will register it. First, the examiner must find that the proposed mark functions as a source identifier — meaning consumers would actually rely on the celebrity's voice to distinguish goods or services. Second, the mark must be distinctive, either inherently (suggestive or arbitrary) or through acquired secondary meaning under Section 2(f). Third, the specimen of use (a sound file for a sound mark, or a real-world advertisement showing the name in commerce) must demonstrate bona fide use in interstate commerce.

For sound marks specifically, the USPTO requires a digital audio specimen in .wav, .mp3, or .wma format, the date the mark was first used anywhere and in U.S. commerce, and a clear description of what consumers hear (for example, "a female vocal speaking the words '...,' accompanied by background music"). The examining attorney at the USPTO will compare the specimen against any prior registered sound marks in the USPTO database to assess likelihood of confusion. McConaughey's voice trademark, granted in 2024, reportedly took roughly a year to register because the office action process required refinement of the goods description and proof of secondary meaning through advertising spend. Swift's 2025 applications were still pending as of late 2025, with at least one Office Action issued challenging the breadth of the entertainment-services description.

Why Celebrities Are Filing These Applications Now

The timing is not coincidental. Between 2023 and 2025, the commercial quality of generative voice cloning improved sharply; ElevenLabs, Resemble AI, and open-source models such as OpenVoice and XTTS-v2 made it possible to reproduce a recognizable voice from as little as three seconds of reference audio. The Federal Trade Commission received more than 5,400 voice-cloning complaints in 2024, up roughly 300 percent year-over-year, and the FBI issued a public alert in 2024 warning that synthetic audio was being used in grandparent scams. Against that backdrop, intellectual property lawyers began treating trademark filings as a defensive layer — not a complete shield, but a registered right that produces a public record, a presumption of validity, and statutory damages availability under 15 U.S.C. § 1117 in counterfeiting cases.

The Backstreet Boys followed a similar path, filing sound-mark applications covering each member's individual voice as well as the group's harmonized sound. Their stated reason, articulated by counsel in a JD Supra interview, was that voice marks create evidentiary hooks that are easier to enforce abroad through the Madrid Protocol than right of publicity claims, which are jurisdiction-specific. The same calculus explains filings by McConaughey, the Estate of James Earl Jones (for the Darth Vader voice), and several K-pop agencies that filed collective voice marks in 2024 for groups under management.

What a Voice Mark Can and Cannot Do

The strength of a voice trademark is its commercial scope. If the USPTO registers your voice in Class 9 for audio recordings, you can sue a competitor who sells AI-generated narration tracks that use a voice confusingly similar to yours, claiming trademark infringement under 15 U.S.C. § 1114 and false designation of origin under 15 U.S.C. § 1125(a). You can also pursue claims for dilution by blurring or tarnishment under 15 U.S.C. § 1125(c) if your voice qualifies as a famous mark under the six-factor test from Coach Servs. v. Triumph Learning. A 2024 survey by Lumen Studios found that voice familiarity scores above 60 percent among U.S. adults were generally sufficient to support secondary meaning arguments in USPTO responses.

What a voice mark cannot do is stop purely expressive uses. Non-commercial deepfakes, political satire, commentary, and most news reporting are protected by the First Amendment and are not actionable under the Lanham Act. The recent ruling in Andersen v. Stability AI (N.D. Cal. 2024) clarified that a registered mark does not, by itself, create liability for model training on publicly available audio without an additional showing of consumer confusion in the output market. Likewise, foreign servers, non-U.S. defendants, and open-source models distributed without a commercial wrapper remain difficult targets even with a registered mark, because the Lanham Act's "use in commerce" requirement presumes U.S. activity that affects U.S. consumers.

Comparison: Trademark vs. Right of Publicity vs. NO FAKES Act

The following table lays out the practical differences between the three principal legal tools available to a person whose voice or likeness is being cloned without consent.

For most individuals, the practical strategy is layered: a state right-of-publicity claim provides the broadest traditional protection, a federal trademark provides the strongest cross-border commercial remedy, and the NO FAKES Act, when it is enacted, will close the expressive-use gap.

Practical Steps If You Want to File

If you are a public figure, a voice actor, or a creator with measurable consumer recognition, the process starts with a clearance search in the USPTO TESS database and the WIPO Global Brand Database. A prior identical or confusingly similar sound mark in the same Nice Class is a hard blocker, and a similar mark in an adjacent class may force you to narrow your goods description or argue no likelihood of confusion. After clearance, you file a Section 1(b) "intent-to-use" application, pay the base filing fee of $350 per class (or $250 per class for TEAS Plus), and submit a digital audio specimen once you have actual use in commerce.

The realistic timeline from filing to registration is 8 to 14 months for a clean application, longer if you receive an Office Action. According to data published by the USPTO in 2024, sound marks had a first-action allowance rate of approximately 38 percent, compared with 56 percent for word marks, which means most applicants should budget for at least one round of prosecution. Lawyer fees for a single voice mark typically run between $2,500 and $7,500 through registration, not including Office Action responses, which add $1,000 to $3,000 per round. After registration, you must file Section 8 and 15 affidavits between the fifth and sixth year, and renew every ten years, with maintenance costs of roughly $525 to $1,200 per filing depending on the number of classes.

Common Mistakes and Pitfalls

The first mistake is treating the trademark as a complete solution. Practitioners who handle AI misuse cases consistently note that right-of-publicity claims do the heavy lifting in litigation, while trademarks serve as supporting evidence and commercial remedies. A second mistake is overreaching on the goods description. Swift's Class 41 application was challenged because "entertainment services in the nature of recording and producing music" is broad enough to touch every artist in the industry, and the examiner asked for a more limited phrasing. A third mistake is failing to police the mark. Trademarks can be challenged for abandonment if the owner does not enforce against obvious infringers, and a string of unenforced AI clones in the market can be used by a defendant to argue genericness or non-functionality.

A fourth mistake, common among individual creators, is filing in only one Nice Class. Voice actors, for example, often need Class 9 (recorded media), Class 35 (advertising endorsement services), and Class 41 (entertainment performance) to cover the realistic range of AI-cloning use cases. Under-classing leaves gaps that a sophisticated infringer can exploit by routing through an unregulated class. A fifth mistake is ignoring international filings. Because AI models train on global data and serve global audiences, an application through the Madrid Protocol designating the EU, Japan, and South Korea is increasingly standard for anyone whose content reaches beyond the U.S. market.

When the Filing Window Is Worth Prioritizing

There are three concrete situations in which a voice or likeness trademark should be on the top of a legal to-do list rather than on a long-range roadmap. The first is when there is measurable consumer recognition — a Lumen-style survey showing 40 percent or higher unaided name or voice recall within a target demographic. The second is when a specific infringer is already using the voice commercially, because a filed-but-unregistered application can support a temporary restraining order under 15 U.S.C. § 1116 in some circuits, and registration provides statutory presumptions that swing settlement leverage. The third is when the celebrity or creator is about to launch a major campaign or product line, because the registration date becomes a priority date for cross-class enforcement.

Outside those situations, the cost-benefit math shifts. For a creator with a small but growing audience, a $4,000 to $10,000 investment in a federal filing may be premature. State right-of-publicity protection, a strong terms-of-service agreement, and periodic platform takedown requests often provide more protection per dollar until the brand reaches a level where consumer confusion is genuinely plausible.

The Bottom Line for AI Trademark Review Readers

Voice and likeness trademarks are real, they are growing, and they are worth understanding, but they are not a magic shield against deepfakes. They are one instrument in a multi-instrument legal kit, and their real value is the priority date, the public record, the cross-border enforcement pathway, and the statutory damages that kick in when a registered mark is counterfeited. For celebrities, voice actors, and high-recognition creators, the move is to layer a federal trademark filing on top of state right-of-publicity rights, a clear terms-of-service regime, and active monitoring of AI marketplaces. For everyone else, the current right answer is to wait for the NO FAKES Act to either pass or fail, revisit the calculus, and in the meantime rely on platform-level enforcement tools that have improved steadily since 2024.