The Ninth Circuit held in 1988 that deliberately imitating Bette Midler's distinctive voice for a commercial could support a California tort claim. The ad used a licensed song and a different singer, yet the court treated the borrowed vocal identity as a separate legal problem.
That distinction matters because copyright protects fixed creative works such as recordings and compositions, not a person's identity in the abstract. A new performance can avoid copying the protected sounds of an old master and still be designed to make listeners recognize a particular singer.
The split now sits close to the musicians' AI identity lawsuit, which alleges misuse of artist identities rather than ordinary infringement of a particular song. For musicians, that moves the legal focus from ownership of a track toward the commercial use of a recognizable persona.
The important limit is easy to miss. Those cases did not establish a general ownership right over genre, mood, phrasing, production habits, or every performance that reminds listeners of an artist. The Waits court focused on deliberate imitation of his voice rather than mere imitation of his style.
That makes synthetic vocal identity disputes more precise than the loose claim that an AI song simply "sounds like" somebody. A plaintiff still needs a legally protected identity interest under the applicable state law, plus facts connecting the challenged use to that identity and the kind of exploitation the law actually reaches.
Publicity rights also do not erase free-expression protections. Courts have long treated advertising and straightforward commercial exploitation differently from uses with stronger expressive, newsworthy, cultural, or parodic interests, so the context in which an imitation appears can matter as much as the acoustic resemblance itself.
That result does not contradict Midler or Waits. It shows why lawyers care about what was taken. An impersonator producing new sounds can raise an identity claim, while copying the singer's actual fixed recording can push the dispute back into federal copyright territory.
The distinction is highly relevant to generative systems because "voice cloning" can describe several technically different acts. A product might reuse a source clip, transform recorded material, train on recordings, generate entirely new audio, or produce an imperfect imitation without reproducing protected sounds from any one master. Those routes can create different legal theories even when the listener's reaction is simply that the output sounds like the same person.
A 2025 federal ruling involving an AI voice company made that separation unusually concrete. The court allowed state civil-rights claims over alleged synthetic voice clones to continue, while treating copyright claims differently depending on whether they concerned an original recording, training activity, or generated output. It was a procedural ruling, not a final declaration that every unauthorized clone violates publicity law.
Congress still had not enacted a general federal digital-replica right by September 2026. The current NO FAKES proposal had advanced from the Senate Judiciary Committee but remained pending, which means performers cannot assume one nationwide statute supplies the answer when an AI system imitates a voice or likeness.
That leaves an awkward practical divide. Copyright can be powerful when protected lyrics, compositions, or sound recordings are actually copied, but it does not automatically give a singer ownership over every new sound that evokes them. Publicity law can reach identity instead, yet its scope depends heavily on state law and on whether the disputed use fits that jurisdiction's rules.
For artists and AI companies, the factual record therefore matters before anyone reaches for a legal label. Identifying the source recordings, the generated audio, the prompt, the model's use of identity cues, the commercial setting, and whether actual protected sounds were reproduced can determine whether the dispute is mainly about a copyrighted work, a person's identity, both, or neither.
That distinction matters because copyright protects fixed creative works such as recordings and compositions, not a person's identity in the abstract. A new performance can avoid copying the protected sounds of an old master and still be designed to make listeners recognize a particular singer.
The split now sits close to the musicians' AI identity lawsuit, which alleges misuse of artist identities rather than ordinary infringement of a particular song. For musicians, that moves the legal focus from ownership of a track toward the commercial use of a recognizable persona.
AI voice imitation can trigger publicity rights
Midler was not a one-off. In 1992, Tom Waits prevailed after a commercial used an impersonator whose performance was deliberately modeled on his distinctive voice, and the Ninth Circuit described the claim as an invasion of his right to control the commercial use of his identity.The important limit is easy to miss. Those cases did not establish a general ownership right over genre, mood, phrasing, production habits, or every performance that reminds listeners of an artist. The Waits court focused on deliberate imitation of his voice rather than mere imitation of his style.
That makes synthetic vocal identity disputes more precise than the loose claim that an AI song simply "sounds like" somebody. A plaintiff still needs a legally protected identity interest under the applicable state law, plus facts connecting the challenged use to that identity and the kind of exploitation the law actually reaches.
Publicity rights also do not erase free-expression protections. Courts have long treated advertising and straightforward commercial exploitation differently from uses with stronger expressive, newsworthy, cultural, or parodic interests, so the context in which an imitation appears can matter as much as the acoustic resemblance itself.
Imitation and copying create different legal exposure
A particularly useful boundary appears in a later Ninth Circuit case involving singer Debra Laws. Sony used part of an actual copyrighted recording containing her performance, and the court held that her state-law claim was preempted because the claimed use was embodied in the copyrighted sound recording itself.That result does not contradict Midler or Waits. It shows why lawyers care about what was taken. An impersonator producing new sounds can raise an identity claim, while copying the singer's actual fixed recording can push the dispute back into federal copyright territory.
The distinction is highly relevant to generative systems because "voice cloning" can describe several technically different acts. A product might reuse a source clip, transform recorded material, train on recordings, generate entirely new audio, or produce an imperfect imitation without reproducing protected sounds from any one master. Those routes can create different legal theories even when the listener's reaction is simply that the output sounds like the same person.
A 2025 federal ruling involving an AI voice company made that separation unusually concrete. The court allowed state civil-rights claims over alleged synthetic voice clones to continue, while treating copyright claims differently depending on whether they concerned an original recording, training activity, or generated output. It was a procedural ruling, not a final declaration that every unauthorized clone violates publicity law.
Federal law still leaves identity protection unfinished
The federal Copyright Office has itself treated digital replicas as a problem that existing law does not fully solve. In 2024, it recommended a new federal law aimed at unauthorized digital replicas and said protection should extend beyond celebrities because ordinary people can also suffer serious harms from realistic synthetic impersonation.Congress still had not enacted a general federal digital-replica right by September 2026. The current NO FAKES proposal had advanced from the Senate Judiciary Committee but remained pending, which means performers cannot assume one nationwide statute supplies the answer when an AI system imitates a voice or likeness.
That leaves an awkward practical divide. Copyright can be powerful when protected lyrics, compositions, or sound recordings are actually copied, but it does not automatically give a singer ownership over every new sound that evokes them. Publicity law can reach identity instead, yet its scope depends heavily on state law and on whether the disputed use fits that jurisdiction's rules.
For artists and AI companies, the factual record therefore matters before anyone reaches for a legal label. Identifying the source recordings, the generated audio, the prompt, the model's use of identity cues, the commercial setting, and whether actual protected sounds were reproduced can determine whether the dispute is mainly about a copyrighted work, a person's identity, both, or neither.
