By Jea Mo
Paul Lehrman was listening to a podcast in 2023 when he heard something that should not have been possible.[1] He heard his own voice, reading words he had never spoken, for a company he had never worked with.[2] This synthetic voice, marketed under the alias of “Kyle Snow,” was created by the artificial intelligence (AI) company Lovo from recordings Lehrman submitted years earlier to an anonymous researcher who was, in truth, a Lovo employee.[3] Lehrman and fellow voice actor Linnea Sage sued, and the Southern District of New York issued the first real judicial test of how existing law handles AI voice cloning in 2025.[4] The ruling made clear that voice actors currently have almost no federal protection against AI voice cloning.[5]
This holding rests on doctrine created by Midler v. Ford Motor Co., where an advertising agency hired a Bette Midler sound-alike after Midler herself declined to license her voice for a commercial.[6] The Ninth Circuit Court of Appeals held that deliberately imitating a distinctive, widely known voice for commercial purposes is a tortious misappropriation under state right-of-publicity law.[7] The court also noted that this was not a copyright matter at all because a voice itself is never “fixed” or copyrightable material to begin with, as copyright protects recordings rather than the sound of the person.[8] Legal protection for voice actors has rested on this holding for nearly four decades—if a voice was copyrightable, federal copyright law would likely preempt any state-law claim covering the same conduct. Because it is not, state tort law has room to operate.
Federal law offered Lehrman and Sage almost no relief. Their Lanham Act claim failed because their voices functioned as their product rather than as a source-identifying trademark distinguishing that product from a competitor’s.[9] Their copyright claims failed for a related reason: an AI model that learns to mimic pitch, cadence, and timbre without reproducing an actual fixed recording is not infringing anything because copyright in a sound recording does not reach an “independent fixation” that merely imitates the original.[10] Both rulings reflect the reality that federal intellectual property law simply is not built to protect voices in such situations.
What did survive was a claim under New York’s Civil Rights Law. The court held in a footnote that the New York claim is not preempted by the Copyright Act for the same reason Midler was never a copyright case: a right grounded in identity rather than a fixed work does not conflict with the exclusive rights Congress reserved to itself.[11] This holding was not separately briefed, and no other circuit has ruled on how Midler applies to a model trained on a performer’s own recordings, rather than a human sound-alike. A court that treats such a clone as a derivative work, not an independent imitation, could find the state-law claim preempted after all.
The matter of which state’s laws apply can be a decisive factor in a case. New York’s statute reaches only the person or entity that directly uses a voice without consent for advertising or trade.[12] Tennessee’s ELVIS Act, shaped by Nashville’s music industry lobbying, extends liability to distributors of voice-cloning technology, not just to whoever ultimately uses the clone.[13] This implies that a Tennessee plaintiff can potentially reach both the AI company that built the tool and the end user, perhaps even when the end user is anonymous. A voice actor cloned under identical circumstances who has viable claim in Nashville may not find the same to be true in New York, given the patchwork of state law.
SAG-AFTRA—a labor union representing more than 160,000 media professionals, such as actors, announcers, dancers, stunt performers, and more—addressed this gap in its 2025 Interactive Media Agreement (IMA). The IMA’s consent-and-compensation requirements for AI digital replicas do not depend on any court’s reasoning holding up because the protection is contractual, not doctrinal.[14] However, that fix only reaches IMA-covered performers on IMA-covered projects, a category most voice actors do not belong to. A 2026 survey by the National Association of Voice Actors found that 21% of respondents had lost work directly to AI, up from 14% the year before.[15]
The NO FAKES Act currently pending in Congress could create a federal intellectual property right addressing unauthorized digital replicas of an individual’s voice and likeness, largely preempting the state-by-state patchwork Lehrman and Sage had to navigate.[16] Whether Congress passes it, and whether it would replace the state-law patchwork or simply add a layer on top of it, remains unclear. Regardless, Lehrman has already illustrated that existing law can address AI voice clones, only barely, and only for the right voice, the right state, and the right judge.
Jea Mo is a 2L at Vanderbilt Law School from Amherst, MA, and Seoul, South Korea. She graduated from Dartmouth College with a degree in English and Linguistics. She will be joining Hogan Lovells Cadwalader in Boston as a 2027 summer associate.
[1] Lehrman v. Lovo, Inc., 790 F. Supp. 3d 348, 357 (S.D.N.Y. 2025).
[2] Id.
[3] Id.
[4] Id.
[5] Id. at 369.
[6] Midler v. Ford Motor Co., 849 F.2d 460, 463 (9th Cir. 1988).
[7] Id.
[8] Id. at 462.
[9] Lehrman, 790 F. Supp. 3d at 378–79.
[10] See id. at 378.
[11] See id. at 381 n. 23.
[12] See N.Y. Civ. Rights Law §§ 50–51 (2024).
[13] See Tenn. Code Ann. § 47-25-1105(a)(3) (2026).
[14] See SAG-AFTRA Members Approve 2025 Video Game Agreement, SAG-AFTRA (July 9, 2025), https://www.sagaftra.org/sag-aftra-members-approve-2025-video-game-agreement.
[15] The State of Voiceover: 2026 Voiceover Survey (Part Three: Artificial Intelligence (AI)), National Association of Voice Actors, https://navavoices.org/vo-survey-2026/voiceover-survey-2026-part-3-ai/; see also The State of Voiceover: 2025 Voiceover Survey (Part Three: Artificial Intelligence (AI)), National Association of Voice Actors, https://navavoices.org/voiceover-survey-2025/voiceover-survey-2025-part-3-ai/.
[16] See NO FAKES Act of 2025, S. 1367, 119th Cong. (2025).