In the Age of AI, Who Really Owns Who You Are?
Taylor Swift is filing trademark applications to protect her voice. James Earl Jones transferred his vocal rights before he died. Khaby Lame sold his digital twin for nearly a billion dollars. These are not isolated stories — they are different attempts to solve the same structural problem: copyright was written for a world in which replicating an identity required significant effort. That world no longer exists.
In the Age of AI, Who Really Owns Who You Are?
In the space of a few months, some of the most closely followed stories in popular culture have raised the same question without answering it.
Taylor Swift — recently named by the New York Times among the thirty greatest living American songwriters — filed three trademark applications covering her voice and image a few weeks ago. James Earl Jones, the original voice of Darth Vader across all Star Wars films, had transferred the rights to his voice two years before his death. Khaby Lame sold the rights to his digital twin for 975 million dollars in January. Disney signed a billion-dollar deal with OpenAI to bring its characters to an AI platform, then pulled out six months later when that platform shut down. The major record labels sued AI music generation platforms; those with enough negotiating power struck deals, everyone else was left out.
The common problem in all of these cases is one: copyright protects works, not identities. And the law was written for a world in which replicating an identity required significant effort.
Consent Is Not Enough
James Earl Jones, the voice of Darth Vader since 1977, lent his voice to that character for nearly fifty years. In 2022, two years before his death, he transferred the rights to his vocal recordings to Lucasfilm, entrusting the Ukrainian startup Respeecher with cloning his voice so the character could continue to exist after him — a technology first used in the Disney+ series Obi-Wan Kenobi. His family then authorized Epic Games and Disney to use that voice in Fortnite in May 2025. According to the press release, Jones “felt that the voice of Darth Vader was inseparable from the story of Star Wars, and always wanted fans of all ages to be able to continue experiencing it.”
SAG-AFTRA, the American actors’ union representing approximately 160,000 entertainment professionals across film, television, and video games, filed an unfair labor practices complaint the same day. Not because Jones’s consent had been ignored — it had been respected to the letter. The problem was different: for years, other actors had voiced the character in video games and smaller productions, replicating his timbre and rhythm. With AI, that work vanished without negotiation, notice, or compensation. As SAG-AFTRA stated in a press release: “We must protect our right to bargain terms and conditions related to voice uses that replace the work of our members, including those who previously performed the work of replicating the iconic rhythm and tone of Darth Vader in video games.”
Jones’s individual consent and the collective rights of the workers who reproduced that voice professionally were two distinct planes — but AI had caused them to collide in an entirely new way. Commenting on the case for CBS News, a union negotiator summarized the point: “If companies had given every actor the same transparency and compensation they gave James Earl Jones, there would have been no strike. It shows they can do it. They just don’t want to do it for those without enough leverage to negotiate.” Consent, in short, is necessary but insufficient — and worth something only to those with enough power to negotiate its terms.
Me, Her, and the Other
In 2013, Spike Jonze directed Her, a film in which a man falls in love with an AI voiced by Scarlett Johansson. Sam Altman, founder and CEO of OpenAI, publicly stated that the film inspired him “more than a little” and that it is his favorite sci-fi work.
In May 2024, OpenAI launched GPT-4o with a new voice called Sky. Johansson, who had refused an offer from Altman a few months earlier to become the official voice of ChatGPT, declared herself “shocked, angered, and in disbelief” at how closely that voice resembled hers — to the point that her closest friends could not tell them apart. Altman tweeted “her” on the day of the launch; OpenAI withdrew Sky within a week.
The company stated it had not used Johansson’s recordings. That was probably true — and that is precisely the problem. Copyright protects original works fixed in tangible form: a recording, a film, a text. As established by the Midler v. Ford Motor Co. ruling of 1988: “A voice is not copyrightable. The sounds are not ‘fixed’.” The right to one’s commercial identity — technically called the right of publicity — exists in some form in the United States, but varies by state and exists neither as a unified federal right in the US nor anywhere in Europe.
The Midler case remains the most solid precedent for a claim like Johansson’s: singer and actress Bette Midler had prevailed against Ford because an advertising agency had hired a vocal substitute for a commercial after Midler declined to participate. The structure of the cases is nearly identical. That precedent is thirty-six years old.
The No AI FRAUD Act, introduced in Congress in 2024 to create federal protections for voice and likeness, is still under debate. The EU AI Act mandates transparency on deepfake outputs from 2026 but does not resolve the question of identity ownership. The European Parliament voted in March 2026 on a report on copyright and generative AI that pushes toward licensing obligations and transparency requirements for training data — but we are still in the realm of recommendations, not binding law.
In the meantime, Taylor Swift has filed three trademark applications. It is the most advanced strategy available in the absence of better tools: building protections where the law does not yet reach, using instruments designed for entirely different purposes.
The Largest Failed Experiment
In November 2025, Disney CEO Bob Iger announced a billion-dollar deal with OpenAI: Disney, Marvel, Pixar, and Star Wars characters would be made available on Sora, OpenAI’s video generation platform. Users would have been able to create content featuring over two hundred characters, some of which would have appeared on Disney+.
The reaction from creatives was immediate and harsh. Disney held its position, formalizing the deal in December with a distinction its lawyers considered crucial: the license covered outputs, not the training of underlying models. OpenAI would not be permitted to use the Disney catalog to improve its foundational models. A license for creative expression, not for technological development.
In March 2026, OpenAI shut down Sora. Disney withdrew from the deal.
Disney’s new CEO Josh D’Amaro, at the annual shareholder meeting, said: “Creativity at Disney is always human-led, and always will be. Our goal with AI is to augment human creativity, not replace it.”
Reducing this Disney story to yet another case of bold entry and rapid retreat misses the point. There is something more complex and more interesting at stake: a demonstration that even the company with the most powerful intellectual property catalog in the world has not yet found a formula that works. And that the distinction between “licensing for output” and “licensing for training” has become the fault line of the entire industry: on one side, those who want to use content as raw material to improve their models; on the other, those who want to control exactly what is generated and how. Two incompatible business models pretending to negotiate over the same object.
Is the Music Over?
In June 2024, the RIAA — the American recording industry association representing Universal, Sony, and Warner — sued Suno and Udio, two of the leading AI music generation platforms, for unauthorized use of protected recordings in training their models. The accusation was direct: building those systems had required copying decades of the world’s most widely heard music.
Over two years, the situation has fragmented in a revealing way. Warner Music settled with Suno in November 2025, signing an agreement that includes a licensing collaboration. Universal Music settled with Udio in October 2025, negotiating royalties per generation (between $0.002 and $0.005 per output, with higher rates for commercial use) and verification rights over training data. Sony is the only major still in litigation: a crucial fair use ruling is expected in July 2026 and could redefine the rules for the entire industry.
The major labels had enough negotiating power to sit at the table and set terms. The same cannot be said for independent artists. Their class action against Suno, filed in California in November 2025, is still open and represents thousands of musicians and small labels whose catalogs were used in training without consent or compensation. The settlements reached by the majors do not extend to them.
Untangling the Knot
All the cases described here share a common structure: someone lost control over something that belonged to them. Not because someone clearly violated a law, but because the law had not anticipated that replicating an identity could become simple.
Copyright was written to protect works — as established by the Midler ruling of 1988, cited in nearly every subsequent legal analysis of AI voice cloning cases. A voice is not copyrightable: sounds are not “fixed.” What is at stake is more personal than any authored work. A character is protectable, but only insofar as it is fixed in specific recordings — not as an abstract entity that can be replicated at will.
Trademark protects distinctive signs associated with products or services. Swift is attempting to use it to protect her voice: the smartest move available in the absence of better tools, not a structural answer to the problem.
The right to one’s identity exists in partial and fragmented forms. For Khaby Lame to transfer the rights to his digital twin, he had to build a contractual framework that law firm Herbert Smith Freehills described as “largely uncharted territory,” sitting “at the intersection of image rights, copyright, trademark, data protection, and AI regulation — many of which have not yet been settled either by statute or case law.”
The problem, in other words, is not that the law does not work. It is that the law was written for a world in which replicating an identity required significant effort. That world has not existed for approximately three years.
How We Approach This at Rational AI
These cases involve real people and fictional characters. But the structural problem they raise — who controls what can be replicated, on what basis, with what governance — applies to any organization with its own digital assets: data, models, processes, knowledge bases, intellectual property embedded in systems.
In this context, “Your Data, Your AI” is not just our payoff. It is the practical answer to the same question that Swift, Lame, Disney, and the major labels are each trying to solve in their own way, often late and under pressure. The difference is that their solutions are available only to those with enough negotiating power to use them. For everyone else, the only alternative is building governance before the problem arrives.
If you want to know more, send us a message.
Originally published on Geckosoft website, translated with Sonnet 4.6 through Rational AI.