Taylor Swift is trademarking phrases to blunt AI voice clones. The rest of us don't have that shield.

Taylor Swift filed trademark applications covering the phrases "Hey, it's Taylor" and "Hey, it's Taylor Swift" earlier this year, with Matthew McConaughey separately seeking trademark protection for some of his most recognizable catchphrases. On the surface the moves look like a celebrity counterstrike against voice AI. The mechanism underneath is narrower: trademark protects brand identifiers, not the person.

For the roughly 330 million Americans who are not Taylor Swift, the protection the law actually offers is a patchwork of state right-of-publicity statutes, federal bills that have not passed, and platform consent rules that depend on enforcement nobody is funded to do. The technology, meanwhile, has dropped to minutes. AI voice cloning tools can recreate a person's voice from just a few minutes of audio, and consumer platforms like ElevenLabs and Suno have made synthetic voices frictionless to produce.

A podcast clip, a YouTube interview, a voicemail greeting, or a customer-service call is technically all the source material a cloning system needs today. The celebrity version of voice protection is a brand-asset argument. The non-celebrity version is a legal scavenger hunt.

Trademark law was built to stop consumer confusion in commerce, not to protect a person. Swift's filings cover specific phrases tied to specific recordings; the protection extends to those phrases, used in commerce, in ways likely to confuse a listener. A clone of her voice saying "the rumors are false, I'm not getting married" probably does not infringe. A clone of her voice saying "Hey, it's Taylor, welcome to my surprise drop" is a different case. McConaughey's catchphrase filings follow the same template, according to Northeastern Global News. Neither approach gives the celebrity a general right to control synthetic versions of their voice. It gives them a defensible position in a narrow lane.

If your voice shows up in a podcast, a YouTube interview, a customer-service call, a TikTok, or a leaked voicemail, it is already in the cloning zone. The question is what the law does about it, and the short answer is that it depends on which state you live in.

Twenty-eight states recognize a right of publicity in some form, and the details vary by jurisdiction. Some cover voice explicitly, some cover likeness only, some require proof of commercial harm, some do not. Tennessee's ELVIS Act (Ensuring Likeness Voice and Image Security) is the first state law to name voice and likeness together in the AI context, and it is the closest thing the country has to a voice-protection template. The statute, as summarized by a third-party regulatory tracker, previews the kind of language other state legislatures are likely to copy. But "closest thing" is doing work in that sentence. Coverage turns on where you live, whether you can afford to litigate, and whether a court treats a synthetic voice as a likeness under that statute's wording. For most people, the cost of bringing a case is the case itself. The framework exists; the access does not.

A cluster of federal proposals, the NO FAKES Act and its siblings, would create a national right against unauthorized digital replicas and put liability on platforms that host them. The bills have been reintroduced in successive Congresses and have not passed. As long as they remain proposals, a voice cloned in California and hosted on a server in Virginia is governed by whichever state's law a plaintiff can afford to argue. State law gives you a case-by-case right, if you can pay for it. Federal law would give you a uniform right, if it existed. Neither is the same as a right that works by default.

Major platforms have begun deploying audio provenance tooling, the kind of C2PA-style content credentials (a tamper-evident watermark standard for media provenance) that mark synthetic audio as synthetic. RVCBench, a recent preprint, benchmarks how robust modern voice cloning systems are against provenance detection, and the results are uneven. The watermark is a defense in depth, not a lock. Voice cloning vendors also publish benign use cases, localization, audiobook production, and accessibility content, which is real and which is also, per one IP-law analysis, exactly the framing the law has not yet decided to regulate.

The narrowest honest forecast is also the most useful: the next two years will produce the first ELVIS Act court test, the first NO FAKES Act hearing in the new Congress, and a measurable shift in how many major platforms label synthetic audio by default. None of those events will eliminate the gap. Each of them will tell the rest of us whether the celebrity shield is becoming a national one, or a hereditary one.