In what amounts to a first for a prominent music artist, Taylor Swift’s company has submitted trademark applications for several aspects of her voice patterns and likeness, effectively attempting to own AI training data legally.
Three trademark application were submitted (through TAS Rights Management LLC) on April 26th, which include specific phrases such as "Hey, it's Taylor" and an image taken during one of her concerts; these are among the first applications using a new legal tactic to address areas within copyright law that protect artists from being synthesized into voice and/or image generation systems.
These trademark submissions come about roughly four months after synthetic versions of Taylor Swift songs began appearing on Tik Tok, garnering in excess of 30M views until eventually being removed. These incidents echo those seen with the 2023 Heart on My Sleeve controversy, in which an AI generated version of a Drake & Weeknd collaboration briefly reached No.1 on popular streaming playlists. In contrast to the 2023 Heart on My Sleeve controversy, Swift is seeking to protect the source material of an individual’s identity rather than the musical composition – a difference that creates potential legal precedent.
Each of the applications also outline coverage for downloadable and recorded verification software for verifying authentic Swift content – it is possible that Swift’s team expects there will be some form of technology that fans can utilize in order to differentiate between real and artificial representations of Swift.
Traditional copyright law protects original creations once they have been put into a tangible format however it does not protect a person’s voice or physical appearance. Trademark law was created to prevent consumers from becoming confused regarding whether or not they are purchasing an authentic product versus a counterfeit product. Given the failure of copyright law to evolve at a rate comparable to technological advancements – it is likely that individuals representing artists will look to utilize existing laws as creatively as possible.
According to James Grimmelmann, Professor of Digital and Information Law at Cornell Tech, who reviewed the filings “We’re seeing artists try out all sorts of different ways to employ legal mechanisms.” Trademark law “isn’t really meant for” this type of application “but when copyright law doesn’t adapt to the evolution of technology, folks start getting creative with whatever options exist.”
Get the latest model rankings, product launches, and evaluation insights delivered to your inbox.
Entertainment services, including personal appearances by celebrities and voice work for animated characters are covered under the definitions of services included in each application. It is conceivable that the court could find that this definition applies to any AI-based voice generator that generates the voice of Taylor Swift for financial gain — although this has yet to be tested in a court of law.
Representatives for Swift did not provide further comment other than stating that the filing process is confirmed. The typical timeframe for review of trademark applications through the United States Patent and Trademark Office is approximately eight-to-twelve months; however due to the novelty of certain types of claims, reviews may take longer than average.
It appears that the timing of these filings is intentional. In early March 2024, Tennessee (Swift’s home state) enacted the ELVIS Act – which provides civil liability for entities utilizing unauthorized AI voice replication. Although federal legislation aimed at establishing similar protections exists, it remains stuck in committee. Additionally, although the NO FAKES Act was introduced in July 2019 — establishing national personality rights — there has been considerable pushback from tech-related interests who argue that passage of the bill would potentially stifle legitimate AI research.
Synthetic versions of Taylor Swift continue to proliferate. There is a Discord server with 47K users that share new generations of Taylor Swift performing music ranging from death metal to corporate jingle renditions. The most advanced examples use Anthropic’s voice transfer model; however Anthropic prohibits creation of any content without explicit permission from the company.
Even though enforcing trademark rights for voice may prove problematic, artists may still need to submit defensive trademark applications. Social media platforms are experiencing growing pressure to create audio verification tools prior to regulatory mandates requiring them. Companies developing voice synthesis products should anticipate litigation that tests the limits of fair use. Fans may soon need tools to verify whether they are consuming actual artist content or synthetically produced content. This could serve as a precursor to a wave of trademark filings by entertainment professionals and creators alike.
By late December, the Patent Office will issue its first formal response regarding the applications; likely requesting additional explanation regarding why certain voice patterns qualify as trademarks. Swift’s team will then have three months to provide an answer. Once trademark rights are granted, enforcement becomes a separate issue. Trademark owners must continually monitor usage and enforce their rights against violators or lose protection.
If every pop star files trademark applications for their unique voice characteristics, the U.S. Patent and Trademark Office faces a dilemma it was never intended to solve – managing biometric signature registrations. For example, clearing a karaoke rendition may now require searching a voice pattern registration database. Weddings singers may need license agreements permitting them to sing like the original. When you consider that a singer’s instrument is their vocal cords, the distinction between flattery and infringement become increasingly blurred.