A. The Landmark Ruling in Lehrman v. Lovo Inc.

(1) Facts and Issues: Unauthorized Cloning of Voice Actors' Voices

One day in 2023, Paul Lehrman, a voice actor working in New York, stopped in his tracks while listening to a podcast.

It was a podcast called Deadline Strike Talk, co-produced with MIT. The narration sounded like his own voice. But he had never recorded it.

He went home and started searching the internet. He found a company called Lovo, a startup offering AI voice synthesis services. On the company's website, he discovered an AI voice listed under the name 'Kyle Snow.' He pressed play.

It was his voice.

Lehrman traced his memory. In 2020, he had taken a job through the freelance platform Fiverr. The client said it was for "academic research purposes." For "internal testing only." The client promised it would never be used commercially or in broadcasts. Lehrman recorded the script, collected his payment, and forgot about it.

Four years passed. That "research" recording had become training data for an AI model. His voice had been cloned and was being sold through Lovo's subscription service. For a few dollars a month, anyone could create content using the 'Kyle Snow' voice. Ads, audiobooks, YouTube videos, podcasts. All without Lehrman's permission.

Another voice actor had a similar experience. Linnea Sage. She too had been deceived the same way on Fiverr in 2019. Her voice was being sold under the name 'Sally Coleman.' Lovo had even produced a promotional video placing Sage's real voice and the cloned version side by side. It was meant to impress investors at the Berkeley SkyDeck Demo Day. "Look how precise our technology is." That was their pitch.

The two voice actors hired lawyers. On May 16, 2024, they filed a class action lawsuit in the U.S. District Court for the Southern District of New York. The complaint listed sixteen causes of action: copyright infringement, trademark violations, New York right of publicity claims, breach of contract, and consumer protection law violations.

The core issue in this case looked straightforward.

When someone clones a person's voice using AI and uses it commercially, which laws apply?

The answer was not straightforward at all.

Is a voice protected by copyright?

Does it function like a trademark, identifying the source of goods or services?

Or does it belong to an entirely different legal category?

On July 10, 2025, Judge J. Paul Oetken issued a sixty-page opinion. It became one of the most detailed legal analyses of voice rights in the AI era.

The judge dismissed some claims and allowed others to proceed. The dividing line was telling. Federal law offered the voice actors almost no protection. State law was their last line of defense.

This ruling matters for a reason. AI voice cloning technology is already widespread. A few minutes of recorded audio is enough to clone anyone's voice. But the law has not caught up. The Lehrman v. Lovo decision exposed both the limits and the possibilities of the current legal framework.

(2) Right of Publicity and the Lanham Act

The voice actors' lawyers looked to federal law for hope.

The Lanham Act is the federal trademark statute in the United States. Section 43(a) prohibits "false endorsement" and "false advertising." In plain terms, you cannot make it look as though someone else is endorsing your product when they are not.

The lawyers' argument went like this: when Lovo sold cloned voices under the names Kyle Snow and Sally Coleman, consumers might mistakenly believe that the actual voice actors had approved the service. That constitutes false endorsement.

Judge Oetken rejected this argument. The reason lay in a basic principle of trademark law.

What trademark law protects is the "source-identifying function." When you see the Nike swoosh, you know it is a Nike product. When you see the golden arches, you know it is McDonald's. The question was whether Paul Lehrman's voice served that same function.

The judge drew a distinction. A celebrity's voice and an ordinary voice actor's voice are different things. When Morgan Freeman narrates something, people recognize it: "Oh, that's Morgan Freeman." His voice identifies the source. But Paul Lehrman is different. He is a talented voice actor. He has appeared on TV programs like Blue Bloods, New Amsterdam, and The Resident. Still, the general public does not hear his voice and think, "That's Paul Lehrman."

The judge wrote: "Plaintiffs use their voices in a manner that is clearly separated from their identity or persona. Their clients purchase recordings of plaintiffs reading scripts and use those recordings in content production."

This is a subtle but critical distinction. A voice actor's voice is the product itself.

It is not a badge of origin.

When people choose the Kyle Snow voice, they think, "I like how this voice sounds," not, "Paul Lehrman made this, so the quality must be good."

The judge posed a thought experiment.

If the plaintiffs' logic were accepted, what would happen when a newcomer with a coincidentally similar voice tried to enter the industry? Could Lehrman file a trademark suit against that newcomer? The Lanham Act has no requirement of intent or bad faith. A mere likelihood of confusion is enough. That leads to absurd results.

The copyright claims were also largely dismissed.

The key provision here is 17 U.S.C. Section 114(b). This section states that copyright in a sound recording does not extend to "the making or duplication of another sound recording that consists entirely of an independent fixation of other sounds, even though such sounds imitate or simulate those in the copyrighted sound recording."

Put plainly, it works like this.

If you record a song, that recording is protected by copyright. But if someone else imitates your performance and makes a new recording, that is not copyright infringement. The copyright in the underlying composition is a separate matter. But "making your voice sound similar" is not, by itself, infringement.

AI does exactly this. Lovo's AI analyzed Lehrman's original recordings.

Pitch, volume, timbre, tempo, intonation, breathing, roughness, vibrato, overall clarity. It learned all of these characteristics. Then it generated a new voice that mimicked them. The judge was explicit: "Copyright law does not protect a voice itself. It does not protect imitation or simulation. It protects only the direct reproduction of a fixed recording."

Only one copyright claim survived. Lovo had used Sage's original recordings directly in investor presentations and YouTube promotional videos. That fell outside the scope of the license. It violated the promise that the recordings would be used only for "academic research purposes."

The limits of federal law were laid bare. Copyright did not apply. Trademark law did not apply. So where could voice actors find relief?

(3) Scope of Protection Under the New York Civil Rights Law

The answer lay in state law. Sections 50 and 51 of the New York Civil Rights Law.

This law establishes the "right of publicity." The right of publicity is the right to control the commercial use of one's own name, likeness, photograph, and voice. This right belongs to the individual. Using it without permission is unlawful.

Lovo pushed back. The New York Civil Rights Law had recently been amended to add protections for "deceased individuals' digital replicas." Lovo's argument ran as follows.

The fact that legislators deliberately added a provision for deceased persons means the existing law did not cover digital replicas of living persons. If the original law already protected the living, there would have been no reason to create a separate provision for the dead.

Judge Oetken rejected this logic. The addition of the deceased persons provision was not intended to narrow existing protections. It was intended to expand them. Living persons were already protected. The amendment extended that protection to the deceased.

The judge emphasized the purpose of the New York Civil Rights Law. This law protects individual identity. Whether alive or dead, a person's voice is part of their identity. An AI-generated voice clone constitutes a "recognizable reproduction" of that identity.

An important legal principle was recognized: the "continuing violation" doctrine. Lovo argued in its defense. We trained our AI model in 2020. The lawsuit was filed in 2024. The statute of limitations has expired.

The judge saw it differently. AI voice cloning is not a one-time act. Every time a Lovo customer generates content using the Kyle Snow voice, a new violation occurs. The statute of limitations begins to run anew with each new use.

The practical significance of this principle is enormous. Once an AI model is trained, it operates for years. Without the continuing violation doctrine, the limitations period would begin running from the moment training occurred. Victims could lose their opportunity to sue before they even know they have been harmed. The continuing violation doctrine prevents this unjust outcome.

The breach of contract claims also survived. Lovo's agents contacted voice actors through Fiverr. They said it was for "academic research purposes." They promised "no commercial use." These conversations were recorded electronically.

Lovo pushed back. Under the Statute of Frauds, contracts must be in writing. Fiverr messages are not formal contracts.

The judge applied a modern interpretation. Section 5-701 of New York's General Obligations Law provides that electronic communications can satisfy the "writing" requirement. Negotiation through Fiverr, agreement on terms, payment of consideration. All of this constitutes a binding contract.

The consumer protection claims were also sustained. Sections 349 and 350 of the New York General Business Law. The allegation that Lovo made false representations about the scope of use for the recordings. This encompasses a broader range of misrepresentations than the Lanham Act.

The lesson of Lehrman v. Lovo is clear. Remedies for AI voice cloning are difficult to find in federal intellectual property law. Voices are not subject to copyright protection. An ordinary person's voice does not function as a trademark. But state law publicity rights do apply. And contract law still works. Let me put this together. On July 10, 2025, Judge J. Paul Oetken signed a 60-page interlocutory decision.

The judge's conclusion was straightforward. Federal law did not favor the voice actors. But they were not without recourse. The judge wrote: "Claims for misappropriation of voice may be properly asserted under Sections 50 and 51 of the New York Civil Rights Law. Unlike copyright or trademark law, these statutes are designed to balance the interests at stake."

Lovo's attorneys won a partial victory. The federal trademark claims were dismissed. For a voice actor's voice to function as a "trademark," it must identify the source of goods. But Lovo did not use Lehrman's and Sage's voices as source identifiers. Their voices were the product itself. This was a subtle but decisive distinction.

The copyright claims were also largely dismissed. Voice itself is not subject to copyright protection. Sound recordings are protected, but an AI voice that imitates those recordings does not automatically constitute infringement. There was one exception, however. The allegation that Linnea Sage's original recordings were directly used in Lovo's marketing materials and investor presentations. That claim survived.

And the judge left one door open. The claim that using recordings to train an AI constitutes copyright infringement. This claim was dismissed, but with leave to amend. The judge gave the plaintiffs 14 days. To supplement with more specific factual allegations.

Plaintiffs' counsel seized that opportunity.

On July 31, 2025, they filed a Second Amended Complaint. A document describing in greater detail how copyright was infringed during the AI training process. Simply alleging that "recordings were used for training" was insufficient. They needed to explain how copying occurred during training and why that copying did not qualify as fair use.

Lovo's attorney Michael Lazarov needed time. On August 7, he asked the judge for an extension to respond. The judge granted the request. The new deadline was September 15. Lovo could file its answer or a new motion to dismiss by then. Meanwhile, discovery continued.

In American civil litigation, discovery is the phase where both sides can examine each other's evidence. Document production requests, interrogatories, depositions. Hidden truths often emerge during this process. How many voice actors' voices Lovo actually collected. Whether Lovo's claim that Kyle Snow and Sally Coleman had "minimal sales" was true. How the voice actors' voices were used in presentations shown to investors.

On June 26, both parties jointly requested an extension of the discovery deadline. The judge approved it. The deposition deadline was extended to August 15, 2025. On the same day, a Stipulated Protective Order was entered. An agreement on how to handle confidential materials disclosed during discovery. This signaled that both sides had sensitive information at stake.

The case carried the potential to expand into a class action.

The plaintiffs claimed to represent two classes. The first was the "Voice Actor Class": all voice actors whose voices were cloned by Lovo. The second was the "Consumer Class": all customers who purchased Lovo software and used the misappropriated voices. Class certification had not yet been decided. But the judge ruled that as long as the individual claims survived, the class action claims survived alongside them.

Let us calculate the distance to a jury trial.

According to the initial Case Management Plan established on August 12, 2024, the expected trial length was six days. But no one knew when that trial would take place. Fact discovery, expert witness discovery, additional motions to dismiss, summary judgment motions. All of these stages remained ahead. Complex civil litigation in U.S. federal court typically takes two to three years to reach trial. This case was filed in May 2024. Trial would come no earlier than late 2026, possibly 2027.

And even after trial, it is not over. The losing side can appeal to the Second Circuit Court of Appeals. Another one to two years in that process. If it reaches the Supreme Court, even longer.

On October 31, 2025, the last document was entered on the court docket. As of January 2026, the case remains ongoing. There is an irony here.

Lovo argued in court that Kyle Snow and Sally Coleman were "unpopular and had minimal sales." It also claimed to have "voluntarily removed them from the platform." But the plaintiffs dispute this. The two AI voices were still being promoted on Lovo's website. If Lovo truly believed these voices were worthless, why not remove them? And why wage such a prolonged legal battle over voices that supposedly have no value?

Consider the legal fees. At major New York law firms, partner-level attorneys charge over $1,000 per hour. Drafting a 60-page opposition to a motion to dismiss takes hundreds of hours. Reviewing documents during discovery and preparing for depositions takes hundreds more. Combined, both sides have likely spent hundreds of thousands of dollars on this litigation, perhaps millions.

Lovo pitched investors at the 2020 Berkeley SkyDeck Demo Day. Displaying Linnea Sage's actual voice alongside the cloned version. Boasting that they could "replicate it perfectly." That presentation was posted publicly on YouTube. To attract investment. And that very presentation has now become courtroom evidence.

There is a reason this case matters.

Judge J. Paul Oetken wrote in his opinion: "This case raises a number of difficult questions, some of which are without precedent. It also carries potentially weighty consequences, not only for voice actors, but for the rapidly growing AI industry, for other holders and users of intellectual property rights, and for ordinary citizens who fear losing control over their own identities."

Federal intellectual property law failed to provide answers for AI voice cloning. Trademark law was powerless when a voice was not a "source identifier" but the "product itself." Copyright law protected sound recordings but not voices themselves. New York state law was different, however. Civil Rights Law Sections 50 and 51. Enacted in 1903, these statutes were being applied to AI voice cloning in 2025. Paul Lehrman heard his own voice on a podcast one day in 2023. A voice he had never recorded. Nearly three years have passed since that moment. He is still fighting in court. No one knows when it will end. No one knows how much he will receive. If he receives anything at all.

One thing is certain. Whatever the outcome of this case, it will set precedent. When AI clones a human voice, how far can the law extend its protection? Where federal law is blocked, can state law open the door? Can a Fiverr message become a legally binding contract? The answers to these questions will emerge from this case.

As of January 2026, those answers have not yet arrived. The courtroom remains open. The clock keeps ticking. Legal fees keep accumulating.

Meanwhile, in Hollywood, a more high-profile dispute was unfolding. The confrontation between Scarlett Johansson and OpenAI.

B. Why Scarlett Johansson Was Furious

(1) The Scarlett Johansson/OpenAI Dispute: The Controversy Over a Voice Resembling the Film 'Her'

On May 13, 2024, OpenAI unveiled GPT-4o. The "o" stood for "omni." A multimodal AI integrating text, image, and voice. At the demonstration, OpenAI introduced a new voice assistant. A female voice named Sky. Warm, natural, and slightly flirtatious in tone.

People watching the demonstration had the same thought. Isn't that Scarlett Johansson?

In the 2013 film "Her," Johansson voiced the AI operating system "Samantha." The male protagonist falls in love with that voice. Spike Jonze's film was hailed as a prophetic work about the relationship between humans and AI. And Sam Altman had publicly said it was his favorite movie.

Immediately after the demonstration, Altman posted a single word on X (formerly Twitter). "her"

Was this coincidence? Or deliberate suggestion?

Scarlett Johansson issued a statement on May 20. According to her account, Altman contacted her in September 2023. He asked her to voice ChatGPT 4.0. Altman told her that her voice could "bridge the gap between consumers and AI." She declined "for personal reasons."

May 2024, two days before the GPT-4o announcement. Altman reached out again. Asking her to reconsider. The demonstration took place before Johansson could respond. She wrote in her statement: "When I heard the demo, I was shocked, angered, and in disbelief that Mr. Altman would pursue a voice that sounded so eerily similar to mine, that my closest friends and news outlets could not tell the difference."

NPR commissioned an independent verification. The voice analysis laboratory at Arizona State University conducted a forensic analysis. Researchers compared the Sky voice against approximately 600 professional actresses' voices. The result: Johansson's voice was more similar to Sky than more than 98% of the comparison subjects. Vocal tract measurements were nearly identical.

OpenAI pushed back. It said Sky's voice belonged to a different professional voice actress. That actress had already been cast before any contact with Johansson. Her identity could not be disclosed for privacy reasons.

But OpenAI suspended the Sky voice on May 19. Altman said in a statement it was "out of respect for Ms. Johansson."

Johansson's lawyers sent two letters to OpenAI demanding a detailed explanation of how the Sky voice was developed. The letters signaled a potential legal claim.

Legally, Johansson's case differs from Lehrman. Johansson is a celebrity. Her voice carries "secondary meaning." People hear her voice and recognize her. That makes a trademark claim under the Lanham Act considerably stronger.

The precedent matters more.

Midler v. Ford Motor Co., 1988. Ford wanted to use Bette Midler's song in a TV commercial. They asked Midler to perform it herself. She refused.

So Ford hired Midler's backup singer to imitate her. The Ninth Circuit ruled in Midler's favor.

Even without direct use of a voice, deliberately imitating a celebrity's voice for commercial gain constitutes a violation of the California right of publicity, the court held.

There is also Waits v. Frito-Lay from 1992.

Singer Tom Waits was famously opposed to having his songs used in advertising.

After Waits turned them down, Frito-Lay hired a sound-alike singer. The court again ruled for the plaintiff.

Johansson's case maps precisely onto these precedents. OpenAI asked her. She refused. Then a voice "eerily similar" to hers appeared. Altman tweeted "her." He was implying the connection.

James Grimmelmann, a professor at Cornell Law School, told CNN: "If OpenAI hadn't spent the last two weeks implying to everyone that 'we just made Samantha from Her,' they might have had a plausible defense."

Johansson ultimately did not file suit. OpenAI had already pulled the Sky voice. She effectively obtained the result of an injunction without needing one. But she raised a broader issue in her statement.

"I look forward to resolution in the form of legislation to help ensure that individual rights are protected."

It was a warning about the lack of legal safeguards against AI voice cloning. Even an A-list actress like Johansson had to deploy lawyers to defend her own voice. What is an ordinary person supposed to do.

(2) Arijit Singh v. Codible Ventures (India)

July 26, 2024. The Bombay High Court. Justice R.I. Chagla issued a historic injunction.

Arijit Singh. One of India's most famous singers. A playback singer on more than 661 Bollywood films. 107 awards. His voice is instantly recognizable across India.

Someone was cloning that voice with AI.

Singh's lawyers explained to the court: a company called Codible Ventures operated an AI platform where users could create new songs in Singh's voice. 456 of Singh's songs had been used as training data.

AI voice cloning was not the only problem. Singh's identity was being used without authorization all over the internet.

Products bearing his name and face were sold on Amazon and Flipkart. Website domains had been registered in his name: arijitsingh.com, arijitsingh.in. Events falsely advertised his appearance.

Justice Chagla was appalled. He wrote in his order: "What shocks the conscience of this Court is the manner in which celebrities, who are performers like the Plaintiff in particular, are susceptible to being targeted by unauthorized generative AI content."

The judge recognized "personality rights" and the "right of publicity" under Indian law, grounding them in Article 21 of the Indian Constitution (the right to life and personal liberty) and civil tort law. He set the scope of protected personality attributes broadly: name, voice, vocal style, vocal technique, vocal arrangements and interpretations, singing mannerisms and habits, even signature.

This is far broader protection than American case law offers. In Lehrman, the U.S. court declined to protect "the voice itself" under copyright. The Indian court recognized "vocal style and technique" as protected aspects of personality rights.

The scope of the injunction was comprehensive. The defendants were barred from using Singh's name, voice, vocal style, technique, photographs, images, signature, persona, or any other aspect of his personality for commercial or personal purposes without his express consent.

The covered media were extensive: physical media, digital media, the metaverse, online platforms, publications, advertisements, merchandise, domain names, generative AI, and voice conversion technology.

This decision is the first judicial ruling on AI voice cloning in India. It may serve as a reference for courts in other countries. The concept of personality rights is recognized in many civil law jurisdictions. The Indian ruling extends that concept to fit the AI era.

The case is still ongoing. An injunction is only interim relief. A final determination will come at trial. But an important message has already been sent: cloning a celebrity's voice and identity with AI carries legal risk.

(3) George Carlin: Posthumous Digital Personality Rights

On January 9, 2024, the Dudesy podcast uploaded a one-hour video to YouTube. The title: "George Carlin: I'm Glad I'm Dead."

George Carlin. A legend of American comedy. A career spanning more than 50 years. A master of social criticism, wordplay, and provocative humor. He died in 2008.

But his voice was heard in the video, commenting on current topics. Reality TV, streaming services, AI technology. All in Carlin's trademark caustic tone.

A disclaimer appeared at the beginning of the video: "I am Dudesy. I am a comedy AI. What you are about to hear is not George Carlin. It is my impersonation of George Carlin, developed in exactly the same way a human impersonator would. I listened to all of George Carlin's material and did my best to imitate his voice, cadence, and attitude."

Carlin's daughter Kelly was furious. She issued a statement: "My father was a legendary comedian, a once-in-a-generation talent. His legacy is the actual body of work he left behind. The claim that AI 'resurrected' him is absurd. The 'George Carlin' in that video is not the beautiful human who raised me with love. I would give anything to spend more time with my father. But no machine will ever replace his genius."

On January 25, the Carlin estate filed suit in federal court in California. The defendants were Will Sasso and Chad Kultgen, hosts of the Dudesy podcast, along with 20 unnamed defendants (5 developers of the AI program, 15 individuals involved in production and sponsorship).

The complaint contained two claims: copyright infringement and violation of the right of publicity.

The copyright infringement claim ran as follows: Dudesy used thousands of hours of Carlin's recordings to train the AI. That constitutes unauthorized reproduction of copyrighted works. The right of publicity claim: California Civil Code Section 3344.1 grants the heirs of a deceased celebrity the right to control commercial use of the person's name, voice, signature, photograph, and likeness. Dudesy commercially exploited Carlin's identity without authorization.

The complaint called the video "computer-generated clickbait" and "a casual theft of the work of a great American artist."

An interesting development followed. After the suit was filed, Kultgen confessed on the podcast. The script was not actually written by AI. A human wrote it. AI was used only for the voice conversion.

That complicated the legal analysis. It weakened the copyright infringement claim; imitating Carlin's "style" is not copyright infringement. But the right of publicity claim remained viable. Replicating Carlin's voice with AI still constituted unauthorized exploitation of his identity.

A settlement was reached on April 3. The terms were straightforward: Dudesy permanently deletes the video, removes all references to Carlin from every platform, and will not use Carlin's image, voice, or likeness in the future without written approval from the estate.

Whether monetary damages were paid was not disclosed. But the estate's attorney, Joshua Schiller, said in a statement: "This settlement will serve as a blueprint for resolving similar disputes in which an artist's or public figure's rights are infringed by AI technology."

The Carlin case demonstrated the importance of posthumous digital personality rights. Deceased celebrities cannot protect themselves. Their voices and identities can be replicated easily with AI technology. Without legal protection, "digital necromancy" will proliferate.

These cases exposed the need for a legislative response. U.S. federal law still contains no provision directly governing AI voice cloning. State laws have begun to fill that gap.

C. The ELVIS Act and What Followed

(1) Tennessee's ELVIS Act: Protecting Postmortem Publicity Rights

March 21, 2024. Robert's Western World, Nashville, Tennessee. A historic bill-signing ceremony took place at this bar, known as a holy ground of country music.

Tennessee Governor Bill Lee picked up his pen. Beside him stood country singers Luke Bryan and Chris Janson. Priscilla Presley was there too. Elvis's former wife.

The bill was called the ELVIS Act. Ensuring Likeness Voice and Image Security Act. The name was chosen, of course, to evoke Elvis Presley.

Tennessee's relationship with the music industry runs deep. Nashville is called "Music City." The Tennessee music industry supports 61,617 jobs and contributes $5.8 billion to GDP. More than 4,500 music venues operate across the state.

And Elvis. He died in Memphis. His legacy is a Tennessee asset.

In 1984, Tennessee enacted the Personal Rights Protection Act. The purpose was to protect Elvis's likeness rights. The law prohibited the unauthorized commercial use of an individual's name, photograph, and likeness.

But AI did not exist in 1984. The voice was not explicitly included as a protected subject.

Forty years passed. In 2023, a fake song called "Heart on My Sleeve," created by AI cloning the voices of Drake and The Weeknd, went viral. The music industry was stunned.

The ELVIS Act introduced two critical changes. First, it added "voice" to the list of protected subjects. The bill defines voice as "a sound in a medium that is readily identifiable and attributable to a particular individual, regardless of whether it is the actual voice of the individual or a simulation of the voice."

This definition matters. It covers not only the "actual voice" but also "simulations." AI-generated voice clones fall within the scope of protection.

Second, it created secondary liability for AI tool providers. The bill established two new categories of liability.

Distributing, transmitting, or making available material with knowledge that it constitutes an unauthorized use of a person's voice or likeness.

Providing an algorithm, software, tool, or technology whose "primary purpose or function" is to produce unauthorized replicas of a specific individual's photograph, voice, or likeness.

The second category is groundbreaking. Companies that build and distribute AI voice-cloning tools can be held liable. Not just end users, but technology providers too.

Penalties for violations were strengthened. Civil lawsuits can seek damages. Willful violations allow treble damages. Criminal sanctions also apply. The offense is classified as a Class A misdemeanor, carrying up to 11 months and 29 days of imprisonment and a $2,500 fine.

The Tennessee legislature passed the bill unanimously. The House voted 93 to 0. The Senate voted 30 to 0. Bipartisan support.

There was opposition. TechNet (representing OpenAI, Google, Amazon, and others) argued the bill was too broad. The Motion Picture Association (MPA) expressed concern that filmmaking depicting historical figures could be restricted.

The bill includes exemptions. Uses protected by the First Amendment (freedom of expression). Fair use. Parody. Satire. Commentary. But the precise scope of these exemptions will be clarified through future case law. The ELVIS Act took effect on July 1, 2024. It is the first state law in America to explicitly regulate AI voice cloning. Other states will follow.

(2) California AB 1836 and AB 2602: Regulating Digital Replicas

On September 17, 2024, California Governor Gavin Newsom signed two bills. AB 2602 and AB 1836. They were a response to Hollywood's demands.

In 2023, SAG-AFTRA (the Screen Actors Guild-American Federation of Television and Radio Artists) went on strike for 118 days. One of the central issues was AI. Actors feared studios would create digital replicas of them and produce films without the original performers.

AB 2602 protects living performers. It takes effect on January 1, 2025.

The law limits the enforceability of contract provisions concerning "digital replicas." A digital replica is defined as "a computer-generated, highly realistic, readily identifiable electronic reproduction of an individual's voice or visual likeness."

The key provision works like this. Even if a contract contains a clause allowing the use of a performer's digital replica, it cannot be enforced unless certain conditions are met.

A "reasonably specific description" of the intended use of the digital replica must be included.

The performer must have received legal counsel or negotiated through a labor union representative.

In plain terms, blanket rights-transfer clauses are void. Clauses like "we may use your digital replica in all future media" no longer hold up. The contract must now specify the purpose and manner of use. And it must be demonstrated that the performer understood and consented. AB 1836 protects deceased performers. It amends Section 3344.1 of the California Civil Code.

This law prohibits the commercial use of a deceased person's digital replica in films, television, video games, audiobooks, sound recordings, and similar media without the consent of the estate.

Exemptions exist. News and public affairs broadcasts. Commentary, criticism, scholarship, satire, and parody. Documentary or historical and biographical portrayals (provided they do not create a false impression of genuine participation). Incidental or fleeting uses.

Violations require payment of the greater of $10,000 or actual damages.

Behind both bills lay years of lobbying by SAG-AFTRA. Union president Fran Drescher said: "The AI protections we fought so hard for last year have now been expanded into California law. As California goes, so goes the nation!"

Tennessee and California. The two most important entertainment states in America have introduced regulations on AI voice and image cloning. Different laws are being written state by state. To address the problems of this "patchwork" regulatory landscape, federal-level legislative discussions are underway.

(3) The NO AI FRAUD Act (Federal Bill)

On January 10, 2024, a bipartisan bill was introduced in the U.S. House of Representatives. The No Artificial Intelligence Fake Replicas And Unauthorized Duplications Act. The NO AI FRAUD Act for short. It was co-sponsored by Representatives Maria Elvira Salazar (Republican, Florida) and Madeleine Dean (Democrat, Pennsylvania).

The bill's purpose is clear. It would grant every American a federal-level property right in their own likeness and voice.

Currently, there is no federal law governing the right of publicity in the United States. Roughly 39 states have their own laws. The scope of protection differs. The available remedies differ. The duration of postmortem protection differs. This inconsistency creates legal uncertainty.

The NO AI FRAUD Act proposes a uniform federal standard.

Its key provisions are as follows.

Every individual possesses a property right in their likeness and voice. This right is an intellectual property right. It is assignable and inheritable. It persists for 10 years after death, regardless of whether the individual commercially exploited it during their lifetime.

It is unlawful to distribute, transmit, or make available an unauthorized digital replica or digital voice replica.

Providing a "personalized cloning service" also triggers liability. This means any algorithm, software, tool, technology, service, or device whose "primary purpose or function" is to produce a digital replica of a specific individual.

Statutory damages are specified. For violations involving the provision of a personalized cloning service: $50,000 per violation, or actual damages plus disgorgement of profits, whichever is greater. For violations involving the distribution of unauthorized replicas: $5,000 per violation, or actual damages plus disgorgement of profits, whichever is greater. First Amendment (freedom of expression) exceptions apply. Courts must balance the individual's intellectual property rights against the public interest in the unauthorized use. Satire, parody, news, documentaries, and similar works may qualify as exceptions.

The bill received enthusiastic support. RIAA (Recording Industry Association of America) Chairman Mitch Glazier. UMG (Universal Music Group) Chairman Lucian Grainge. SAG-AFTRA President Fran Drescher. Even IBM and Disney issued statements of support.

Critics exist. The Foundation for Individual Rights and Expression (FIRE) argues the bill threatens the First Amendment. The scope of protection is too broad, the exceptions too narrow. Legitimate satire and commentary about politicians or public figures could be chilled.

The bill did not pass the 118th Congress. A similar bill, the NO FAKES Act, was introduced in the Senate and remains under discussion. After the Taylor Swift deepfake incident in January 2024, legislative momentum grew.

Federal legislation will take time to complete. In the meantime, state laws serve as laboratories. Tennessee, California, Illinois. Each state's experience will feed into federal lawmaking.

AI voice cloning technology has already spread widely. The law is playing catch-up. As the Lehrman v. Lovo decision shows, existing federal law does not provide victims with adequate remedies. State-level publicity rights and contract law are stopgap measures. But the level of protection varies from state to state.

The need for a uniform federal standard is clear. If bills like the NO AI FRAUD Act or the NO FAKES Act pass, every American will receive equal protection over their voice and likeness. AI companies will also be able to plan their businesses under clear rules.

The voice cloning problem does not stay confined to the domain of publicity rights. Combined with deepfake technology, it becomes a tool for crime. The next chapter examines this darker side.