Trademark legislation has lengthy accommodated nontraditional “sensory” marks—sound, scent, shade, movement—after they perform as supply identifiers. Current filings by Taylor Swift and associated entities spotlight each the promise and limits of this framework within the AI period. I’d argue that sensory-mark doctrine is ill-suited to deal with AI techniques that ingest and reproduce “personhood alerts”—voice timbre, stylistic signats, and biometric-adjacent identification options—and proposes a biometric/personhood mark framework. However I haven’t been capable of give you a motive not to do it, so my sense is that it’s a spot to start out.
Generative AI does one thing the prior digital platforms by no means fairly achieved: it erases the road between consuming a piece and producing one. Streaming, social media, and distribution platforms moved content material sooner and farther, however they largely preserved a primary distinction. You listened, watched, or learn. Should you created a brand new work, you probably did so individually from the distribution platform. Generative techniques collapse that sequence right into a single act. The identical system that ingests a novel, a vocal efficiency, or a lifetime of lyrics can, moments later, produce one thing that tries to really feel prefer it got here from the identical supply.
That shift isn’t just about effectivity; it’s not about transformation; it’s about transmogrification. AI fashions don’t merely index or distribute expressive works—they take in them. Coaching datasets turn out to be uncooked materials from which the system learns patterns of language, tone, construction, and efficiency. Over time, these patterns coalesce into one thing greater than statistical correlation. They turn out to be recognizable alerts—what audiences intuitively perceive because the hallmarks of a specific creator. A voice or guitar tone isn’t just pitch and timbre, however phrasing, perspective, emotional cadence. A author’s fashion isn’t just vocabulary, however rhythm, perspective, and narrative intuition. Generative AI captures these parts and recombines them into outputs that will not be copies within the conventional sense, however can perform as substitutes and flood the marketplace for consideration and that means. And as Choose Chhabria stated in Kadrey v. Meta, No different copyright use “has something close to the potential to flood the market with competing works the best way that LLM coaching does.”
That’s the place the authorized stress begins. And given the forces of evil arrayed in opposition to artists on this context, each protection is vital since Large Tech ignores all of them anyway.
New Logos Filed by Taylor Swift on April 24, 2026:
- “Hey, it’s Taylor Swift” Trademark Software
- “Hey, it’s Taylor” Trademark Software
- Taylor Swift-Likeness Trademark Software
Taylor Swift’s TAS Rights Administration filed three new USPTO trademark functions: two sound marks for brief voice clips, “Hey, it’s Taylor Swift” and “Hey, it’s Taylor,” plus a visible mark describing Swift on stage holding a pink guitar whereas carrying a multicolored iridescent bodysuit and silver boots. The filings seem designed so as to add trademark safety to right-of-publicity and different potential cures in opposition to AI impersonation and deepfakes.
The voice-registration technique is novel as a result of trademark legislation historically protects source-identifying sounds, not a star’s voice as such. A brief spoken phrase in Swift’s personal voice could perform as a model sign, however courts haven’t totally examined whether or not that registration can cease AI-generated vocal imitations that evoke her identification with out copying the precise clip. That makes the submitting each defensive and experimental: it makes an attempt to transform recognizable personhood alerts into trademark belongings. That’s probably a lot broader safety that copyright.
AI labs will plainly be excited about ingesting Taylor Swift recordings as a result of they comprise terribly useful personhood alerts: voice, phrasing, timbre, lyrical cadence, efficiency fashion, and audience-recognition cues. Plus they’ve most likely already ingested all of the TS they will steal with each palms. Trademark legislation is unlikely to achieve the coaching act itself as a result of AI coaching is inside copying, not consumer-facing use of a mark as a supply identifier. Trademark legal responsibility often activates whether or not the defendant used the mark in commerce in a means more likely to confuse customers about supply, sponsorship, or endorsement.
The output stage is totally different. If an AI product makes use of Swift’s registered voice clip, picture mark, identify, or branding to promote or determine items or recommend affiliation, trademark claims turn out to be extra believable. That’s the reason the registrations matter. They might not cease ingestion of songs or recordings, however they provide Swift protectable model belongings exterior copyright within the musical work or sound recording. In different phrases, trademark could not management coaching, however it could strengthen claims in opposition to artificial endorsements, faux promotions, and industrial makes use of of Taylor Swift’s personhood alerts.
Nevertheless, proper of publicity is probably going a greater match than trademark for AI voice and likeness issues as a result of it protects identification itself, not merely source-identifying manufacturers. Trademark asks whether or not customers are confused about sponsorship or affiliation; ROP asks whether or not somebody commercially exploited an individual’s identify, voice, likeness, or persona with out consent (which AI labs clearly have finished). That issues for AI as a result of the hurt could also be identification substitution even when no trademark is copied and no shopper sees a model label. Taylor Swift’s trademark filings could assist police faux endorsements and branded makes use of, however ROP extra immediately reaches unauthorized artificial imitation of her voice, picture, or efficiency persona.
