It is Friday, April 24, 2026, late afternoon at the U.S. Patent and Trademark Office. Three filings land in serial. Two are sound marks: a four-second fragment that says “Hey, it’s Taylor Swift,” and a slightly shorter one that says “Hey, it’s Taylor.” The third is a visual mark, a description so specific it reads like a stage cue: a woman holding a pink guitar with a black strap, in a multi-colored iridescent bodysuit, silver boots, on a pink stage, in front of a multi-colored microphone, purple lights in the background.
Three days later, the filings are in every entertainment column. [S: someone grab the popcorn.]
This is the story of every public-facing person and every company in 2026 with a recognizable surface. It is happening right now at yours.
What artists are actually defending against
AI-generated facsimiles. A synthetic Taylor singing a song she never sang. A synthetic version of her that sounds plausibly like her but isn’t quite identifiable enough to bring inside the perimeter of her legal team. The kind of output that is good enough to confuse listeners, good enough to monetize on streaming services, but not so direct a copy that anyone can shut it down.
That gap is where the next decade of music piracy will live. AI tools can produce a song that sounds like Taylor Swift, in Taylor’s style, without using a single sample from her catalog and without saying her name. Copyright reaches a fixed recording. Right of publicity reaches identifiability. Both miss this gap.
A human artist whose music happens to sound like Taylor is a different question. That is style, not impersonation.
Why the intro-phrase filing is brilliant
The trademark is a watermark.
A federal mark on “Hey, it’s Taylor Swift” doesn’t just stop a deepfake hello. It establishes a positive signal of authenticity. Any track that opens with the registered phrase is presumptively hers. Any track that doesn’t open with it isn’t.
That isn’t how trademark is usually framed. It is how the filing actually functions. Swift can put the registered phrase on her own future releases. Or the production. Or the live show. Listeners come to expect it. The presence becomes the verification. The absence becomes the disclaimer.
The synthetic Swift song that omits the watermark is no longer passing as her. It is, by her own design, officially not her. The synthetic that copies the watermark is straightforward Lanham Act infringement. The middle case, the synthetic that sounds like her without using the intro, is the one she has now reframed as “not me, regardless of what your ears tell you.” [M: trademark, consistent use, public expectation. Watermark.]
It is a creative use of trademark doctrine that very few public-facing people have tried at scale, and the playbook gets sharper every time it runs.
What about cover bands?
The textbook answer is they’re fine. Cover bands operate under public performance licenses (ASCAP, BMI, SESAC) that cover the composition. The cover band signals its own identity in branding and announcements. A reasonable listener understands the performance as homage, not original. No source confusion. No § 43(a) violation.
The interesting cases are at the edges.
A cover band that records a studio version of a Swift song that opens with the registered phrase and uploads it to Spotify with metadata that doesn’t clearly identify the cover band is a different problem. The intro on the streaming track creates source confusion. The streaming algorithm could surface the cover next to authentic Swift tracks. The trademark gives Swift a takedown lever the composition license does not.
A synthetic cover band, AI-generated, playing a Swift song with a synthetic version of the intro phrase, is straightforward Lanham Act infringement. No license. No real performers. The intro is the whole basis for source attribution.
A live performance where the band’s lead singer says “Hey, it’s Taylor Swift” before launching into a cover is performative quotation. The trademark does not reach it, for the same reason a comedian impersonating a CEO on stage doesn’t violate the CEO’s trademark. [S: the trademark catches the synthetic, not the local cover band.]
The pattern: the mark reaches uses that confuse a reasonable listener about source. Licensed cover, performative quotation, and clearly-attributed homage are outside it. Synthetic facsimiles using the watermark are inside it. That is exactly the line the filing draws.
Why this is not a music story
It would be reassuring if this were a Taylor Swift story. It is not.
The CMO has a face on a Dropbox of 82 webinars. The CS manager has a voice on 1,500 Gong calls. The VP of Sales has a deal-memo skill on a personal Claude account. The senior litigator has a brief-drafting skill with 40 briefs.
Each of them is at the same fork. AI-generated facsimiles, plausible but not quite identifiable, are coming. The question is whether they have a watermark of their own.
A registered phrase. A registered visual cue. Some signal of authenticity that can be present in real output and absent from synthetic output. Without it, everything that surfaces under their name is potentially theirs and potentially not. With it, the burden shifts.
The recording artist is the most public version of the problem. She is also the first to deploy the watermark response.1
If your business uses voice talent or audio IP, the AI-generated equivalents are already being trained on the originals. The contracts need to catch up.
Talk to a Talairis attorney →A closing thought
She already owns her voice. Copyright protects her recordings. Right of publicity protects her likeness. The trademark adds a watermark.
That is the operational move. With the registered phrase on the front, the track is hers. Without it, the track isn’t. Listeners learn to expect the marker, and the absence does the work.
The artist who files first sets the verification standard for everyone who follows. The companies and public-facing individuals who don’t file are the ones whose authentic output and synthetic facsimiles are indistinguishable to the public.
The catchphrase is the watermark. Use it accordingly.
Postscript
The Swift filings landed on April 24, 2026. Matthew McConaughey got there earlier and broader. His holding company J.K. Livin Brands began filing in 2023 and secured eight federal registrations through late 2025 and early 2026, including a sound mark on “Alright, alright, alright” plus marks on persona and likeness elements. McConaughey filed first. Swift made it news. Spoken-voice trademarks have not been tested in court. The first opinion to construe them will set the perimeter for every public-facing person in the country, and for every company that needed a watermark and didn’t think to file one.2
- Swift’s filing got the entertainment-column treatment. The same play is available to anyone with a recognizable signal: a phrase, an image, a sound. Most companies won’t file. The first wave of synthetic facsimiles will tell them why they should have. ↩
- A trademark used as a watermark depends on three things. Registration, consistent use, and public expectation. The first is mechanical. The second is operational. The third takes time. Filing today is what creates the public expectation that runs against the synthetic five years from now. ↩
