AI-driven voice synthesis has collapsed the technological barrier that once made vocal appropriation laborious and detectable—creating fresh tort exposure under right-of-publicity doctrines whose core principles have not materially changed since Bette Midler sued Ford Motor Company.
I. Doctrinal Foundation: Voice as Protected Identity
The right of publicity protects a person's economic interest in controlling the commercial use of her identity. Voice is among its most personal components. Unlike a photograph, a recognizable voice travels through radio, podcast, and streaming without any accompanying visual cue, making unauthorized replication commercially powerful and legally fraught.
The Ninth Circuit's foundational treatment of voice misappropriation comes in Midler v. Ford Motor Co., 849 F.2d 460 (9th Cir. 1988). Ford's advertising agency, having failed to license Midler's performance of "Do You Want to Dance," hired a sound-alike singer instead. The district court granted summary judgment for Ford, concluding no legal principle prevented imitation of Midler's voice. The Ninth Circuit reversed. The court acknowledged that the voice itself is not copyrightable—it cannot be "fixed" within the meaning of 17 U.S.C. § 102(a)—and that California Civil Code § 3344's enumerated list of protected identity elements (name, voice, signature, photograph, likeness) was inapplicable because the defendants used a different singer's actual voice, not Midler's. Nonetheless, the court held that California common law recognizes a tort when "a distinctive voice of a professional singer is widely known and is deliberately imitated in order to sell a product." Midler, 849 F.2d at 463. The holding rests on a property-rights rationale: by analogy to the companion statute protecting deceased persons' voices, common-law rights in one's voice are property rights, and appropriation without consent is a tort.
Four years later, in Waits v. Frito-Lay, Inc., 978 F.2d 1093 (9th Cir. 1992), the Ninth Circuit refined the Midler framework. Tom Waits — whose raspy, idiosyncratic vocal style is commercially distinctive — prevailed on both a voice misappropriation claim and a Lanham Act false endorsement claim arising from a Doritos commercial that employed a Waits sound-alike. The Waits court added an element Midler left implicit: the imitation must be so convincing that "people who were familiar with plaintiff's voice who heard the commercial believed plaintiff performed it." 978 F.2d at 1108. That threshold keeps the tort from swallowing every impression or parody. Waits also clarified that the California right is not subject to copyright preemption because it protects an aspect of identity—not the fixed expression—and the elements of the tort diverge from copyright infringement.
Together, Midler and Waits establish the operative test in the Ninth Circuit: (1) the plaintiff possesses a distinctive, widely known voice; (2) the defendant deliberately imitated or replicated it; (3) for commercial purposes; and (4) without consent. The deliberateness element has always implied intentional copying, but AI voice cloning makes deliberateness trivially easy to prove—the training data, synthesis prompt, and output all document a purposeful replication.
II. Statutory Landscape: From California Common Law to the ELVIS Act
Most state right-of-publicity statutes predate generative AI and were not drafted with voice synthesis in mind. Florida's Fla. Stat. § 540.08 prohibits unauthorized public use of "the name, portrait, photograph, or other likeness of any natural person" for trade or advertising purposes. The phrase "other likeness" has been read broadly by Florida courts, but the statute does not expressly list "voice." Practitioners litigating AI voice claims in Florida should plead both the statutory claim and the common law misappropriation theory to avoid a textual gap argument from defense counsel.
Alabama enacted a modern right-of-publicity statute in 2015 codified at Ala. Code § 6-5-770 et seq.. The statute expressly includes "voice" in its definition of "indicia of identity" under § 6-5-771, alongside name, signature, photograph, image, likeness, and "a substantially similar imitation of one or more of those attributes." The final phrase—"substantially similar imitation"—maps precisely onto AI-generated voice clones and provides textual support for claims that would otherwise require common law gap-filling. The Alabama statute extends post-mortem protection for fifty-five years, a significant advantage for estates of deceased musicians whose voices might be cloned for commercial exploitation.
Tennessee moved first among states to legislate specifically against AI voice replication. The Ensuring Likeness, Voice, and Image Security Act — the ELVIS Act, Tenn. Code Ann. § 47-25-1101 et seq., effective July 1, 2024 — amends Tennessee's Personal Rights Protection Act of 1984 to expressly prohibit unauthorized commercial use of any individual's voice. The ELVIS Act defines "voice" to include "a sound in a medium that is readily identifiable and attributable to a particular individual, regardless of whether the sound contains the actual voice or a simulation of the voice of the individual." Tenn. Code Ann. § 47-25-1101(4). That definition is clearly directed at synthetic replication: a voice clone that never contains a single millisecond of actual recorded audio nonetheless falls within its ambit. The Act creates a novel secondary liability provision for any person who distributes an algorithm, software tool, or service "the primary purpose or function of which is the production of a particular, identifiable individual's" voice, with knowledge the use is unauthorized. This secondary liability provision reaches AI platform operators and is more expansive than the common law framework from Midler and Waits, which governs only direct appropriators.
III. Applying the Framework to AI Voice Cloning
Several structural features of AI voice synthesis pressure-test the existing doctrine.
Training-data appropriation. A voice synthesis model trained on recordings of a particular individual without license may itself constitute the tortious act, or at minimum a preparatory step toward it. The Midler tort requires that the voice be "deliberately imitated in order to sell a product." If the AI company trains the model to produce a celebrity's voice for commercial API licensing, the commercial purpose element is satisfied at the training stage. Practitioners should consider pleading voice-cloning-as-misappropriation at the model-training level, not just at the point of end-user deployment.
Output authenticity standard. Waits requires that the imitation be convincing enough that listeners believe the artist actually performed. Modern neural voice synthesis routinely satisfies this test — academic research has documented sub-2% detection error rates in voice conversion models. Discovery in any voice-cloning case should include expert testimony on detection failure rates and consumer survey evidence, mirroring the likelihood-of-confusion analysis in Lanham Act trademark cases.
Preemption. Copyright preemption under 17 U.S.C. § 301 remains the most significant threshold defense. Following Midler, state voice misappropriation claims survive preemption so long as the plaintiff's voice — not a fixed performance — is the protected interest. Where the defendant trained the AI on a licensed sound recording and argues that the license extends to synthetic output, practitioners should focus the pleading on the identity-in-voice right, not on the copyrightable expression in the underlying recording.
First Amendment. Satirical AI voice impressions used in clearly non-commercial commentary likely enjoy First Amendment protection. The transformative use test developed in right-of-publicity cases — whether the defendant's work merely exploits the plaintiff's identity or adds something beyond it — will govern. A political parody podcast that synthesizes a politician's voice commenting on legislation presents a far stronger First Amendment defense than a voice-cloned celebrity used in a commercial jingle.
IV. Practice Notes
Jurisdiction selection. Where facts permit, Tennessee offers the most favorable statutory scheme for voice-cloning claims because of the ELVIS Act's express definition of simulated voice and its secondary liability provision. California common law under Midler/Waits remains potent for claims involving music industry figures. Florida and Alabama provide backup statutory theories but with narrower textual bases for voice specifically.
Defendants. Plaintiffs should map the full chain: the AI voice synthesis platform, the API customer who incorporated the voice clone, the media outlet or advertiser that disseminated it, and any music distributor or streaming service that monetized the output. The ELVIS Act's distribution-of-tool provision captures platform operators even where they played no role in the final commercial use.
Temporary restraining orders. Irreparable harm in voice-misappropriation cases is relatively easy to establish because commercial dilution of a distinctive voice — especially in markets for endorsement, audiobook narration, or music — does not admit of easy monetary remediation. Courts in the Ninth Circuit have issued TROs in right-of-publicity cases on proof of deliberate commercial appropriation alone.
Damages. Under Waits, both compensatory and punitive damages are available. Compensatory damages should be measured by market license value (the fee a reasonable licensee would have paid), advertising revenues attributable to the cloned voice, and any diminution in the plaintiff's licensing market. Expert testimony on market rate for voice licenses—common in music industry litigation—transfers to these cases with minimal adaptation.
Pleading specificity. Courts hostile to right-of-publicity claims sometimes require specific identification of the commercial purpose served by the appropriation. Identify with precision in the complaint: the product or service advertised, the medium of distribution, the revenues attributable to the commercial, and the markets in which the plaintiff's voice has been commercially exploited.
V. Open Questions
The right-of-publicity doctrine is conceptually well-suited to AI voice cloning, but several doctrinal gaps remain unresolved.
Who sues when the cloned voice belongs to a deceased artist? Most state statutes extend post-mortem protection; the ELVIS Act and Alabama § 6-5-770 et seq. are explicit. But conflicts arise when the decedent's estate is domiciled in a state with no post-mortem right (some states still limit post-mortem protection to celebrities who commercially exploited their identities during life) while the AI deployment occurs in a state with robust post-mortem rights. Practitioners should perform a careful choice-of-law analysis before filing.
Does federal regulation preempt state law? No federal right-of-publicity statute exists. Proposed federal legislation has been introduced in multiple Congresses but none has passed. The Defend Our Performers and Creators Act (a/k/a "NO FAKES Act") would create a federal right against unauthorized AI-generated replicas, but as of the publication of this article it has not been enacted. Until then, the patchwork of state law governs.
Is a voice clone a "likeness" under statutes silent on voice? Florida § 540.08's "other likeness" language is an open question. One plausible reading treats likeness as limited to visual representations given the statutory enumeration (name, portrait, photograph), but a more purposive reading extends it to any characteristic rendering identity recognizable—which a voice clone indisputably does. Expect early threshold litigation on this textual issue in Florida courts.
VI. Closing
AI voice cloning is not technologically new in principle — the deliberate imitation of a celebrity's distinctive voice to sell products has been tortious in California since Midler in 1988. What is new is the scale, accessibility, and precision with which synthetic voices can now be produced. The doctrinal framework is adequate to handle most commercial voice-cloning claims under existing law, but practitioners need to think carefully about training-data liability, secondary liability for AI platforms, and the First Amendment envelope around expressive uses. State legislators — led by Tennessee — are moving quickly to close textual gaps. Federal legislation may eventually rationalize the patchwork; until then, forum selection and careful claim-threading remain essential.
Talk to Yates Anderson
If you are litigating a matter in this area — or weighing whether to — the working analysis above only goes so far. Request a case evaluation and a Yates Anderson attorney will respond within one business day.
Informational only. Not legal advice. No attorney-client relationship is created by reading this post. Consult a licensed attorney in your jurisdiction.