Facts
Paul Lehrman and Linnea Sage are voice-over actors in New York. In 2019 and 2020 they were hired through the freelance platform Fiverr, by users who according to the complaint were acting for Lovo, to record scripts they were told would be used only for internal research or tests; they were paid $1,200 and $400. Lovo sells a text-to-speech subscription service built on its “Genny” AI software. The actors allege that Lovo cloned their voices, offered the clones to subscribers as “Kyle Snow” and “Sally Coleman”, used them in marketing and an investor presentation, and used the recordings to train its model. They brought a putative class action under New York law, the Lanham Act and the Copyright Act.
Question
On Lovo's motion to dismiss, the court had to decide whether federal trademark and copyright law protect a person's voice against AI cloning, and whether Sections 50 and 51 of the New York Civil Rights Law, which bar using a living person's “voice” for advertising or trade without written consent, cover digital replicas. Lovo argued that Section 50-f, added in 2021 and expressly covering digital replicas of deceased performers, implied that replicas of living persons fall outside the statute, and that the one-year limitation period had run. The court also considered state consumer protection and contract claims.
Decision
Judge J. Paul Oetken granted the motion in part. The Lanham Act claims were dismissed. Copyright protects particular recordings, not a voice as such, so claims based on the voices failed; Sage's claim over use of her original recording survived, and the training-based claims were dismissed with leave to amend. The Civil Rights Law claims survived: following New York case law treating computer-generated images as a “portrait”, the court read “voice” to include digital replicas and found Section 50-f no bar. Because the clones could keep producing new clips, the republication exception could make the claims timely, and fictitious names did not avoid liability. The consumer protection and contract claims proceed; the other common-law claims were dismissed, mostly as preempted.
Why it matters
With the copyright and trademark claims largely failing, the opinion points to state publicity law as the main route for voice-cloning claims. It treats a clone that can generate new speech indefinitely as a continuing use for limitation purposes, and holds that attaching invented names to cloned voices does not avoid liability. These are rulings at the pleading stage on allegations presumed true; liability has not been established. Whether training a model on copyrighted recordings infringes copyright was left open.
Related stages
The opinion and order is dated 10 July 2025 (ECF No. 45). The plaintiffs filed a second amended complaint on 31 July 2025, and Lovo again moved to dismiss the training-based copyright claim. On 27 May 2026 Lovo filed a Chapter 7 petition in the U.S. Bankruptcy Court for the Southern District of New York (No. 26-11249), and the case was stayed before that motion was decided. For other decisions on AI imitation of a person's voice, see yin-ai-voice-beijing and yapay-zeka-ses-klonlama-gorevli-mahkeme.