Voice actors win round one against an AI voice company in New York: what Lehrman v. Lovo means for singers

Two New York voice actors who say an AI company sold clones of their voices have cleared the first hurdle in court. On 10 July 2025, Judge J. Paul Oetken of the Southern District of New York granted in part and denied in part Lovo, Inc.'s motion to dismiss the class action brought by Paul Lehrman and Linnea Sage.

"Beat" would overstate it. At this stage the court takes the complaint's allegations as true and asks only whether they state a legal claim. Nothing has been proven, and Lovo hasn't been found liable for anything. What the 60-page opinion does show is which legal tools work against a voice clone and which don't. For singers, that map matters more than the result.

What do the actors say happened?

According to the amended complaint, as the opinion summarises it, anonymous users on the freelance marketplace Fiverr hired both actors for voice recordings. Lehrman was told his recordings were for internal, academic research only and was paid $1,200. Sage was told hers were test scripts for radio ads that wouldn't be released externally, and was paid $400. The actors say those buyers were a Lovo employee and a Lovo co-founder.

They say they found out when they heard a podcast, Deadline Strike Talk, narrated partly by an AI voice that sounded like Lehrman. Lovo had allegedly been selling clones of the two actors under the names "Kyle Snow" and "Sally Coleman", and had used Sage's real recording in an investor pitch posted on YouTube. In its reply to their cease-and-desist letter, Lovo said its use was "fully permissioned" (opinion, p. 7).

What survived, and what didn't?

The judge kept these claims alive (opinion, pp. 59–60):

• New York Civil Rights Law, Sections 50 and 51, the state's statutory right of publicity.
• New York General Business Law, Sections 349 and 350, on deceptive practices and false advertising.
• Breach of contract, based on the Fiverr messages limiting how the recordings could be used.
• One copyright claim: Sage's claim over Lovo's use of her actual recording in its marketing.

These were dismissed: both federal Lanham Act claims (false association and false advertising), the copyright claims over the clones themselves, contributory infringement, fraud, unjust enrichment, conversion and unfair competition. The copyright claims over training Lovo's "Genny" model were dismissed with leave to amend, because the complaint never explained what training is or how it copies anything. Loeb & Loeb's case summary reports the same split.

Why did copyright and trademark fail?

Copyright. Section 114(b) of the Copyright Act says a new recording that only imitates or simulates the sounds in a protected recording doesn't infringe it. The clones made new recordings that sounded like the actors; they didn't copy the actual sounds the actors recorded. The court said the actors were in effect asking for copyright in "their voices qua voices", and that a voice is too abstract to be copyrighted. It cited a Second Circuit case separating the recognisable sound of 50 Cent's voice, which copyright doesn't cover, from the recordings it appears on, which it does (pp. 42–43).

Trademark. A voice can act as a mark only when it identifies the source of a product. Here, the court said, the voices were the content being sold. It added that letting any performer sue soundalikes under trademark law would create a cause of action for, in effect, plagiarism (pp. 26–28).

Why did New York's publicity law work?

Section 50 bars using the "name, portrait, picture, likeness, or voice of any living person" for advertising or trade without written consent, and Section 51 lets that person sue. Lovo argued that the law doesn't cover AI replicas of living people, because in 2021 New York added a separate provision covering digital replicas of deceased people. The judge disagreed. New York's highest court had already held in 2018, in Lohan v. Take-Two Interactive, that a computer-generated image can be a "portrait" under the statute, and he saw no principled reason to read "voice" more narrowly (pp. 46–49).

Three details matter for musicians:

• Fake names don't help. Selling the clones as "Kyle Snow" and "Sally Coleman" didn't defeat the claim. The court said allowing that would carve a massive loophole into the statute.
• A clone keeps publishing. Section 51 claims must be brought within one year. The court let the claims proceed anyway, because a clone that keeps producing new clips in the actors' voices may count as fresh publication.
• Users can be on the hook too. The consumer-protection claim survived because Lovo told subscribers they got full commercial rights. The court said Lovo was right about federal copyright and trademark law but wrong about New York law, under which its subscribers, like Lovo, could be liable (pp. 53–54).

What we think

For singers, the right that counts is the one attached to the person, not the recording. Many working singers don't own their masters. If the strongest claim against a voice clone ran through the copyright in the recordings, the label would hold it. This opinion points the other way. Copyright won't stop a clone that never reproduces your actual takes, but New York's statute protects your voice as part of your identity, and only you can give the written consent it requires. That's our reading of one district court's reasoning. It isn't binding on other courts, but it is the clearest map yet of where a voice claim lives.

The catch is geography. The US Copyright Office's 2024 digital replicas report called state law a patchwork: not every state protects voice, some protect only well-known or commercially valuable voices, and they typically apply only to commercial uses. It called for a federal law. New York's statute is broad on exactly these points: it names "voice" outright and covers "any living person", famous or not. Tennessee's ELVIS Act, signed in March 2024, added "voice" to that state's law. A singer in a state with neither has a much harder case.

The contract claim is the practical lesson. Lehrman's and Sage's best facts are a few chat messages limiting how their recordings could be used. Anyone selling vocals, toplines or session parts online should get the permitted use in writing, and should say whether AI training or voice modelling is excluded. Fiverr's default terms gave Lovo arguments; the specific messages are what kept the contract claims alive.

The second-order effect lands on the clone marketplaces. The consumer-protection theory turns a platform's "commercial rights included" promise into a liability if the voice wasn't licensed, and it tells the platform's customers they may be exposed too. If that holds, provable consent becomes a feature voice-model sellers have to show, not a footnote. Services built on consented, paid voice deals benefit. Services selling vaguely sourced "royalty-free" voices lose.

Two cautions. The training question is open: those claims were dismissed for thin pleading, with an invitation to replead, and the court noted that Lovo may yet argue fair use. And a right of publicity only bites on commercial use, so it does less against non-commercial deepfakes.

The same theory is now being tested on music generators. A later class action led by Jason Isbell relies on state publicity law instead of copyright, including New York's.

Sources

• US District Court for the Southern District of New York, Lehrman v. Lovo, Inc., Opinion and Order, No. 24-CV-3770 (JPO) (10 Jul 2025)
• US Copyright Office, Copyright and Artificial Intelligence, Part 1: Digital Replicas (Jul 2024)
• Office of the Governor of Tennessee, Gov. Lee Signs ELVIS Act Into Law (21 Mar 2024)
• Loeb & Loeb, Lehrman v. Lovo, Inc. (Jul 2025)