Executive Summary
- What is new: A recent decision from the Southern District of New York, in Lehrman & Sage v. Lovo, Inc., addresses the intersection of AI voice cloning technology and intellectual property rights, focusing on contract law, copyright law and state law protections.
- Why it matters: The decision highlights the legal challenges surrounding AI-generated content, including the enforceability of electronically negotiated contracts, copyright limitations, and the adaptability of New York’s statutory protections for personal identity.
- What to do next: The treatment of using third-party content to train AI models remains an evolving area of the law. Companies should monitor further developments as cases make their way through the courts and new laws and regulations emerge.
__________
A recent 60-page decision from the U.S. District Court for the Southern District of New York, in Lehrman & Sage v. Lovo, Inc. (J. Oetken, 24-CV-3770), addresses the rapidly evolving intersection of artificial intelligence (AI), voice cloning technology and intellectual property rights. The case, brought by professional voice actors Paul Lehrman and Linnea Sage, raises significant questions about the legal protections available to individuals whose voices are cloned and commercialized without their consent. This detailed breakdown of the court’s analysis in each area demonstrates the complex interplay between contract law, intellectual property law and state law protections in the context of AI-generated content and personal identity. The case is another important building block in the jurisprudence surrounding the use of third-party content as AI training data without consent.
Key Points From the Decision
Breach of contract. The court took a broad view of contract formation, acknowledging the way parties transact in today’s world. For example, the court held that electronic messages, including those using screen names or pseudonyms, can satisfy the Statute of Frauds’ writing and signature requirements.
Lanham Act. In dismissing the plaintiffs’ Lanham Act claim, the court found that misrepresentations about the scope of rights a defendant possesses are not “false advertising.” The court also found that the use of a voice clone itself does not satisfy the Lanham Act false association elements where there is no use of the voice to endorse or promote a product or service.
Scope of copyright protection. The court dismissed the plaintiffs’ infringement claim based on the use of their voices to train Lovo’s model; the crux of many AI training data cases today. However, the court indicated that its dismissal was based on the plaintiffs’ failure to provide the appropriate factual allegations about how the training occurred, and granted them leave to amend. The court also dismissed plaintiffs’ infringement claims based on the AI-generated outputs that “cloned” their voices. The court held that the Copyright Act protects only the original sound recordings (the fixed expression), not the abstract qualities of a voice or new recordings that merely imitate or simulate the original.
New York’s right-of-publicity laws (N.Y. Civil Rights Law §§ 50 and 51). The court denied the defendant’s motion to dismiss the plaintiffs’ §§ 50 and 51 claims, holding that plaintiffs had established the unauthorized use of voices for the purpose of advertising or trade. Importantly, the court rejected Lovo’s argument that these statutes have a one-year statute of limitations, finding that since the Lovo model was trained on the plaintiffs’ voices, it was arguably continuing to replicate those voices each time the model generated new clips using their voices.
New York consumer protection laws (§§ 349 and 350). The court also denied the defendant’s motion to dismiss the plaintiffs’§§ 349 and 350 claims, finding that plaintiffs had adequately pled that Lovo had misrepresented the scope of the rights it had. While this was the same argument the court had rejected with respect to plaintiffs’ Lanham Act claims, the court explained that the types of misrepresentations covered by the New York law were broader than what the Lanham Act permitted.
Background
Paul Lehrman and Linnea Sage, professional voice-over actors based in New York, were approached via the freelance platform Fiverr and asked to provide voice recordings for what they were repeatedly assured were internal, academic or test purposes, with explicit assurances that their recordings would not be used for commercial or broadcast purposes. The individuals approached them using pseudonyms but were later revealed to be employees or founders of Lovo, Inc., a company offering AI-driven text-to-speech and voice cloning services.
After delivering the recordings and receiving payment, Lehrman and Sage discovered that Lovo had used their voices to create AI-generated “clones” marketed under the names “Kyle Snow” and “Sally Coleman.” These cloned voices were incorporated into Lovo’s subscription service, used in marketing materials and made available to Lovo’s customers for commercial use. The plaintiffs alleged that their voices were used without proper authorization or compensation, and that Lovo’s representations to both them and its customers were false and misleading.
Lehrman and Sage brought a putative class action against Lovo, asserting 16 claims, including for breach of contract and violations of the Lanham Act, the Copyright Act and New York’s right-of-publicity laws.
Court’s Holding
We highlight below the key holdings from the court’s decision.
Breach of Contract
Lehrman and Sage alleged that they entered into contracts with Lovo’s agents through a series of online communications, which included explicit restrictions on the use of their voice recordings. The court found these allegations sufficient to plausibly establish the existence of enforceable contracts under New York law, which requires offer, acceptance, consideration, mutual assent and intent to be bound. The court therefore denied Lovo’s motion to dismiss the breach-of-contract claim.
The court’s analysis is notable for its recognition of modern contracting practices, especially in the context of online platforms like Fiverr. Specifically, the court held that electronic messages exchanged via Fiverr, including those using screen names or pseudonyms, can satisfy the Statute of Frauds’ writing and signature requirements. The court emphasized that the substance of the agreement, not the formality of signatures or real names, is what matters, especially given contemporary business practices.
The court also found that the essential terms — namely, the nature of the recordings, the price, and the restrictions on use — were sufficiently definite. The court explained that the parties’ mutual assent was evidenced by the explicit back-and-forth regarding the scope of use, and the subsequent delivery and payment for the recordings.
The court also considered Fiverr’s terms of service as part of the contractual framework, noting that these terms could supplement the parties’ direct communications and provide further detail about the rights and obligations of each side.
Lanham Act (False Association and False Advertising)
The plaintiffs asserted two bases of liability under Section 43(a) of the Lanham Act: unfair competition/false association and false advertising. A false association claim requires (1) a showing of a mark that is distinctive as to a source of goods and (2) likelihood of confusion. The court held that because “mark-like devices” such as a celebrity’s image or identity can satisfy the first prong of this test, there was no reason an individual’s voice could not also satisfy this prong. However, the court distinguished between cases where a person’s image, likeness or persona functions as a trademark (such as in celebrity endorsement scenarios) and the present case, where the plaintiffs’ voices serve as the product itself, not as a source identifier. The court reasoned that while a celebrity’s likeness or persona can sometimes serve as a mark, a professional voice used in commercial recordings does not inherently indicate source or sponsorship (i.e., it is not itself a brand, but rather a service to which a brand might be attached).
The court also dismissed plaintiffs’ false advertising claims under the Lanham Act. Under the Lanham Act, a plaintiff must establish either that the challenged advertisement is literally false or that it will mislead consumers. The court held that Lovo’s alleged misrepresentations (e.g., claiming to have “full commercial rights” to the plaintiffs’ voices or using fake names for plaintiffs’ cloned voices) did not satisfy either of these criteria for Lanham Act purposes. The court noted that there was no actionable consumer confusion, as Lovo’s marketing made clear that the voices were AI-generated clones, not the actual voices of Lehrman and Sage, and a false statement about having the appropriate licensing rights is not “false advertising.” The court also found that even if Lovo had engaged in false advertising, plaintiffs failed to allege any injury from such actions.
Copyright Claims
The plaintiffs asserted three direct infringement claims against Lovo:
- Sage asserted that Lovo directly infringed her copyright by using her actual voice recording in marketing materials.
- Both plaintiffs claimed that Lovo infringed their copyrights by using their original voice recordings to train its AI model.
- Both plaintiffs asserted that the AI-generated voice clones constituted infringing derivative works of their voices.
The plaintiffs also alleged that Lovo was secondarily liable for contributory copyright infringement by making the AI-generated voice clones available to third parties, who could then use them in their own projects.
Scope of Lovo’s License Rights
The court first analyzed whether Lovo’s use of the plaintiffs’ recordings was authorized under the relevant contracts and Fiverr platform terms (through which Lovo had acquired their voice recordings). The court found that the Fiverr terms only granted buyers a limited license for noncommercial use, and while buyers can pay for commercial use rights, Lovo did not do so. The court also found that the communications between the plaintiffs and Lovo’s agents (via Fiverr messages) imposed even narrower restrictions — explicitly limiting use to “academic research purposes only.” The court concluded that that Lovo’s use of the recordings was plausibly outside the scope of any license granted, and thus potentially infringing.
Infringement Analysis
- The court denied Lovo’s motion to dismiss on the claim that it used Sage’s actual voice recording for marketing materials and demonstration purposes, reasoning that the scope of the license granted to Lovo clearly did not permit this use.
- The court dismissed plaintiffs’ infringement claim with respect to the use of the plaintiffs’ voice recordings to train Lovo’s AI model, but with leave to amend. The court held that there was insufficient factual detail in the complaint regarding how the AI training process allegedly infringed the plaintiffs’ exclusive rights, but that it would be straightforward for plaintiffs to amend their complaint to make the appropriate allegations. In a footnote, the court noted that Lovo asserted in a single sentence that its training was fair use, but that if plaintiffs amended their complaint, and defendants again moved to dismiss, a more thorough fair-use defense would need to be articulated.
- The court also dismissed the plaintiffs’ claims that the AI-generated voice clones were an infringing derivative work of their actual voices. The court held that copyright law protects only the original sound recordings (the fixed expression), not the abstract qualities of a voice or new recordings that merely imitate or simulate the original. The Copyright Act expressly excludes from protection new recordings that are independent fixations of other sounds, even if they closely mimic the original.
- The court also dismissed plaintiffs’ contributory infringement claim, concluding that since the AI-generated outputs did not constitute direct infringement of the original recordings, there could be no contributory infringement.
New York Civil Rights Law §§ 50 and 51 (Right of Publicity/Misappropriation of Voice)
In order to sustain a claim under New York Civil Rights Law (NYCRL) § 50, a plaintiff must establish the unauthorized use of their name, image or voice for the purpose of advertising or trade. The court found that plaintiffs pled each of these elements, and rejected each of Lovo’s motion-to-dismiss arguments:
- The court agreed with Lovo that §§ 50 and 51 have a one-year statute of limitations, but noted that here, since the Lovo model was trained on the plaintiffs’ voices, it was arguably continuing to replicate those voices each time the model was run.
- Lovo argued that a 2021 amendment to the NYCRL protecting digital replicas of deceased persons meant it did not protect living persons. The court rejected this argument, finding that the statute was amended to add deceased persons, because there was consensus that living persons were already covered by the law.
The court also stated that when confronted with new technologies such as AI, it is important when engaging in statutory construction “to encompass future developments and technological advancements,” and that excluding digital clones from the law would frustrate its statutory purpose.
New York Consumer Protection Laws (§§ 349 and 350)
In order to sustain a claim under New York Consumer Protection Laws (§§ 349 and 350), a plaintiff must allege material misrepresentations. Here, the court found that plaintiffs had adequately pled that Lovo had misrepresented the scope of rights it had (i.e., “Lovo’s subscribers received a product that was worth less than what Lovo promised”). This was the same argument that the court had rejected with respect to plaintiffs’ Lanham Act claims, but the court explained that while Lanham Act misrepresentations were limited to “nature, characteristics, qualities or geographic origin,” the New York law did not include this limitation. The court also found that plaintiffs adequately alleged injury based on the lost sales of their actual voices due to Lovo’s misrepresentation that it held all necessary rights to offer its AI clones.
Conclusion
Overall, the court’s opinion affirms the enforceability of modern, electronically negotiated contracts, clarifies the limits of federal intellectual property statutes in the context of personal attributes, and highlights the adaptability and strength of New York’s statutory protections for personal identity in the digital age. As evidenced by the court’s opinion, the issue of AI-generated voice cloning raises a number of unique issues compared to cases where the AI training was done on text or images and the goal was not to replicate the original. The Lovo decision will therefore be important precedent in cases where AI models seek to replicate some aspect of the identity of an individual.
This memorandum is provided by Skadden, Arps, Slate, Meagher & Flom LLP and its affiliates for educational and informational purposes only and is not intended and should not be construed as legal advice. This memorandum is considered advertising under applicable state laws.
