The rule most people reach for is that a voice cannot be owned. That is correct as far as it goes: copyright protects fixed works, not the sound of a person, and nobody holds a copyright in how they talk.
The exception is a body of law that has nothing to do with copyright, that varies by state, and that has been used successfully against exactly this conduct before.
Scarlett Johansson said on Monday that a voice in OpenAI’s new assistant was eerily similar to her own, that the company had approached her in September about providing a voice and she had declined, and that she had instructed lawyers. The company said the voice belongs to a different professional actress speaking naturally, and paused its use.
The widespread reading is that this is a copyright story with an unclear answer. The record shows it is a right of publicity story with a reasonably clear one.
What the common reading gets right
Copyright genuinely does not help here, and the reasoning is sound.
Copyright subsists in particular recordings and particular performances. It does not subsist in a timbre, an accent, a cadence or a manner. If it did, every impressionist would be an infringer and every singer with a recognizable style would be able to stop anyone who sounded like them.
So a synthesized voice that resembles somebody, built without using any of their recordings, infringes no copyright. That is the answer to the question most people are asking, and it is the wrong question.
What the record shows instead
The relevant doctrine is the right of publicity, which protects a person’s name, likeness and, in several states, voice against commercial appropriation.
It is a state law right, strongest in California and New York for obvious reasons, and it is not about who made the sound. It is about whether a commercial product is trading on somebody’s identity.
The controlling case is nearly forty years old. Bette Midler sued Ford over a commercial in which, after she declined to participate, an agency hired a former backing singer and instructed her to sound like Midler. No Midler recording was used. The Ninth Circuit held that deliberately imitating a distinctive voice to sell a product was actionable.
Read the facts of that case against the facts asserted here and the parallel is close enough to be uncomfortable: an approach, a refusal, and then a product with a voice that sounds like the person who refused.
Which brings up the part that turns a resemblance into a case, and it is not the audio.
A right of publicity claim of this kind turns on appropriation, which is about purpose. Two voices sounding alike is not actionable; choosing a voice because it sounds like somebody, to trade on the association, is.
Which means the strongest evidence is never the waveform. It is the approach in September, the second approach reported to have come days before the demonstration, and anything in the record indicating that the resemblance was the objective rather than a coincidence.
The company’s position, that the voice is a different actress speaking naturally, is a complete defense to copyright and an incomplete one here. Ford’s singer was also a real person singing in her own voice.
What the company did, and what it means
It paused the voice rather than defending it, within days, while maintaining that it had done nothing wrong.
That combination is the standard posture of a party that believes it would probably win and does not want to find out. Litigating a right of publicity claim means discovery, and discovery in a case about intent means internal communications about why a voice was chosen.
The second case, which is even closer
Midler is the famous one. The closer one on the facts is Waits.
Tom Waits sued Frito-Lay over a commercial featuring a singer imitating his distinctive delivery, after Waits had refused to participate. The Ninth Circuit upheld a substantial award, and the case established that the doctrine covers voice imitation for commercial purposes as a matter of settled law rather than as an extension of it.
The account given this week tracks those facts in every respect that mattered in either case: an approach, a refusal, a product, and a resemblance that the company says is coincidental.
Why this is the beginning of something rather than an incident
The general problem is that synthesis has made resemblance cheap, and the law that governs resemblance was built for a world in which producing one required hiring a person to imitate somebody.
In that world the imitation was rare, deliberate and documented, which is why an intent-based doctrine worked. A product that can generate a voice resembling anybody, at no cost, breaks the assumption that resemblance implies choice.
And the doctrine is state law, which means fifty answers, several of which do not protect voice at all, applied to products distributed everywhere at once.
The part of this that is not about one company
The voice actors who do the underlying work are the group with the most at stake and the least leverage, and their position is worse than the famous claimant’s.
A recognizable star has a right of publicity worth asserting and lawyers to assert it. A working voice actor whose sessions are used to train a system that then replaces them has no distinctive identity to protect in law, because the thing taken is not their identity. It is their labor, and the instrument for that is a contract they signed before anybody knew what the recordings would be used for.
That is the genuinely unresolved question underneath this week, and it will not be answered by a dispute between two large parties who can both afford to settle.
It is worth adding that the company had a straightforward way to avoid all of this, which was to have the actress it says recorded the voice say so publicly, with her name attached, at launch. That it did not is not evidence of anything; it is simply the reason a factual dispute exists at all.
Where this actually lands
Not in a courtroom, probably. A claim that is strong enough to be worth bringing and embarrassing enough to be worth avoiding is a claim that settles quietly.
The change will happen inside legal departments instead. The lesson available to every one of them this week is not that voices are protected; it is that approaching somebody, being refused, and then shipping something that sounds like them creates a record. The safest version of this product is the one where nobody was ever asked, which is a strange thing for a week of this kind to have taught.
