The Short Answer: Not Illegal, But Legally Risky

Here's the answer upfront: making an AI voice of someone is not automatically illegal — but using it commercially without that person's permission can land you in serious legal trouble. The difference between "legal experiment" and "lawsuit" comes down to consent and how you use the voice.

Why does this matter right now? Because the barrier to cloning a voice has collapsed. According to legal analysis of voice cloning regulations, today's AI tools can replicate a voice with remarkable accuracy using just seconds of audio. A short clip from a podcast, voicemail, or social video is enough raw material.

And misuse isn't hypothetical. Documented incidents already include AI-generated robocalls mimicking political candidates, synthetic voices used in financial scams, and unauthorized celebrity voice clones. In the lead-up to Australia's 2025 federal election, residents of Burnie, Tasmania were played an AI-generated voice clone of Senator Jacqui Lambie — a real-world case of voice cloning entering political communication.

The legal risk concentrates around a few key frameworks:

• Right of publicity — U.S. courts treat distinctive voices as protectable identity, and California courts have found liability for voice imitations even when no actual recordings were used, if the voice is recognizable and used commercially.
• Contract law — licensing terms bind AI companies even when agreed through informal electronic messages; exceeding the licensed scope counts as infringement.
• Emerging statutes — Tennessee's ELVIS Act is already law, and the federal NO FAKES Act (S.1367, introduced April 9, 2025) would create a nationwide IP right in voice replicas.
• Consumer protection laws — state statutes like New York's §§349/350 can apply where companies misrepresent the rights they hold in a voice.

The court system is already testing these theories. In Lehrman & Sage v. Lovo, Inc., two voice actors sued after their voices — recorded for what they believed was "academic research purposes only" — were cloned and marketed under fake names like "Kyle Snow" and "Sally Coleman." A federal judge in the Southern District of New York allowed the right of publicity and contract claims to proceed, holding that excluding digital clones would "frustrate the statute's purpose."

For businesses, the practical takeaway is simple: legality hinges on consent and context. Legitimate uses — voiceovers, customer service, accessibility for people who have lost their voice — are recognized across jurisdictions when proper authorization exists. That's why at Worqd, our AI voice agents follow permission-aware outreach practices with explicit consent built into every interaction, the same standard we'd recommend any company demand from a voice AI provider.

So while no law says "you may never clone a voice," the combination of publicity rights, contract enforcement, and incoming federal legislation means unauthorized commercial use is one of the fastest ways to turn an AI experiment into a liability.

Which Laws Actually Apply — and Which Don't

Here's the uncomfortable truth: the law that protects your voice the most isn't copyright — it's a patchwork of state statutes most people have never heard of. If you're building AI voice tools into your business, knowing which laws actually bite (and which don't) is the difference between a compliant deployment and a lawsuit.

Right of publicity is the dominant risk. U.S. courts treat distinctive voices as a protectable part of a person's identity, and several states back this with statute: California Civil Code §3344, New York Civil Rights Law §§50-51, and Tennessee's ELVIS Act, which explicitly targets unauthorized AI voice replication. California courts have even found liability for voice imitations when the imitation is recognizable and used commercially — no actual recordings required. The clearest signal came in Lehrman & Sage v. Lovo, Inc., a Southern District of New York case where voice actors alleged their voices were cloned and marketed under fake names. The court allowed the right of publicity claims to proceed, holding that excluding digital clones would "frustrate its statutory purpose."

Copyright, surprisingly, offers little protection. In the same Lehrman decision, the court dismissed derivative work claims, explaining that the Copyright Act protects only the fixed sound recordings — not the abstract qualities of a voice or new recordings that merely imitate the original. Analysts comparing Australian law reach a similar conclusion: AI clones are "novel outputs based on learned patterns," so a substantial-similarity claim is unlikely to succeed. Fair use defenses don't rescue you either — legal commentary argues voice cloning fails the transformative-use test under Andy Warhol Foundation v. Goldsmith because its whole purpose is "to generate voices identical to the input subject."

Contract law, on the other hand, has real teeth. In Lehrman, informal electronic messages — sent through Fiverr, some under pseudonyms — satisfied the Statute of Frauds, and using recordings beyond a licensed "academic research purposes only" scope was found plausibly infringing. Translation: any licensing terms you agree to, however casual, are enforceable.

Two other frameworks round out the picture. Consumer protection claims fared better than Lanham Act claims in Lehrman because the defendant "misrepresented the scope of rights it held" — broader coverage than federal trademark law permits. And on the horizon, the proposed federal NO FAKES Act (S.1367, introduced April 9, 2025) would create a federal right for voice and likeness, require watermarking and disclosure, and establish notice-and-takedown — though it remains just a bill.

For businesses, the practical takeaway is consent and documentation. Before any voice clone touches your customer-facing systems — whether that's an AI receptionist or automated follow-up — you need explicit written permission covering the specific use. It's the same principle we apply across every Worqd engagement: verify what you're allowed to use before you use it, because in voice AI, the paperwork is the law.

• Get explicit written consent covering use, compensation, duration, and exclusivity before cloning any voice.
• Treat informal electronic agreements as binding — document every licensing term.
• Don't rely on copyright or fair use as defenses; courts have rejected both for voice cloning.
• Check state-by-state rules, since protections vary widely across jurisdictions.

Why 'Fair Use' and Copyright Defenses Won't Save You

When people get into legal trouble with AI voice cloning, it's rarely because they ignored the law — it's because they believed the wrong defenses would protect them. Three misconceptions cause most of the damage, and courts are dismantling all of them.

Misconception one: copyright protects voices. It doesn't. The human voice itself is not copyrightable — only fixed sound recordings are. In Lehrman & Sage v. Lovo, Inc., a New York federal court dismissed the plaintiffs' copyright claims outright, holding that the Copyright Act protects original recordings, "not the abstract qualities of a voice or new recordings that merely imitate or simulate the original," according to Skadden's analysis of the decision. Australian legal scholars reach the same conclusion, describing AI clones as "novel outputs based on learned patterns" rather than substantial-similarity copies, per a Wolters Kluwer copyright analysis. Copyright is the wrong tool — for both attackers and defenders.

Misconception two: fair use covers cloning. The Supreme Court's Andy Warhol Foundation v. Goldsmith ruling raised the bar for what counts as "transformative" use, and voice cloning fails that test. The purpose of cloning isn't to extract statistical patterns for new applications — it's to generate a voice identical to the input subject. A Berkeley Technology Law Journal analysis also notes that a viable licensing market already exists for voice rights, which undercuts the fair use market-harm argument entirely. Training on narrow, identifiable datasets tied to specific people is the opposite of transformative.

Misconception three: jurisdiction doesn't matter. It matters enormously. U.S. state laws vary widely — California, New York, and Tennessee have strong statutory protections, while other states offer far less. Key differences include:

• California's 2019 law bans deceptive deepfakes within 60 days of an election, according to legal analysis of voice cloning laws
• New York's Civil Rights Law §§50-51 carries just a one-year statute of limitations — though ongoing replication can extend it
• Australia offers only copyright and passing off claims, leaving major protection gaps
• The proposed federal NO FAKES Act, introduced April 9, 2025, would create a national voice right but preserve state laws existing as of January 2, 2025

Meanwhile, the claims that actually stick come from elsewhere: right of publicity, contract law, and consumer protection statutes. In Lehrman, even informal Fiverr messages counted as binding license terms, and exceeding the licensed "academic research purposes only" scope was plausibly infringing. California courts have found liability for voice imitations even without actual recordings, per Rock LAW's review of synthetic voice regulations.

The practical takeaway: hope is not a compliance strategy. Explicit written consent, documented license scope, and jurisdiction-by-jurisdiction review are the only real defenses. That's why teams deploying AI voice agents — including the work we do at Worqd — build consent verification and clear usage terms into the process from day one, rather than retrofitting them after a demand letter arrives.

How to Use AI Voices Legally: A Consent-First Checklist

The legal ground beneath AI voice cloning is shifting fast, but one principle holds steady: consent is not optional. Courts across California, New York, and Tennessee have affirmed that distinctive voices are protectable identity assets, and the Lehrman v. Lovo decision confirmed that right of publicity claims can proceed even when no original recording is used — only a synthetic replica marketed under a fake name. A federal court in New York allowed those claims to move forward, while dismissing copyright arguments because the law protects fixed recordings, not the voice itself.

That ruling also delivered a practical lesson for anyone licensing voice data: informal electronic agreements bind. Messages exchanged on platforms like Fiverr — pseudonyms and all — satisfied the Statute of Frauds, and using recordings beyond the licensed "academic research purposes only" scope constituted infringement. The court treated those chat logs as enforceable contracts, a reminder that every permission slip matters.

For teams deploying AI voice at scale, a consent-first checklist keeps you on the right side of the law:

• Get explicit written consent covering specific uses, compensation, duration, and exclusivity
• Treat every licensing agreement — even informal messages — as a binding contract
• Verify consent before deployment and monitor ongoing usage for scope creep
• Disclose AI voice use to audiences; watermarking and transparency mechanisms are becoming statutory expectations
• Secure estate authorization for deceased persons' voices — post-mortem publicity rights can last decades

Jurisdiction matters. State laws vary widely, and the proposed federal NO FAKES Act would create a national right while preserving state protections enacted before January 2, 2025. The bill also mandates watermarking and disclosure — requirements that leading providers are already building into their workflows.

At Worqd, we help clients evaluate providers on exactly these compliance practices — because the fastest way to kill a growth engine is a lawsuit that could have been avoided with a better contract. When your AI voice agents qualify leads and book calls under 60 seconds, the last thing you need is a rights dispute slowing you down.

What This Means for Businesses Using AI Voice Agents

If your business uses AI receptionists or voice agents to answer and qualify leads, the legal question isn't whether the technology is allowed — it's whether the voice behind it was properly obtained. The good news: legitimate uses like customer service and voiceovers are widely recognized, and legality hinges on consent and context, according to legal analysis.

That means AI voice agents for inbound sales, after-hours answering, and lead follow-up are entirely defensible when built the right way. The risk concentrates in two places: where the voice came from, and whether callers know they're talking to an AI. California courts have found liability for voice imitations even without using actual recordings when the result is recognizable and used commercially, as IP law firms warn.

The Lehrman v. Lovo decision made the stakes concrete. A New York federal court allowed right of publicity claims to proceed after voice recordings licensed for "academic research purposes only" were cloned and marketed commercially — and held that even informal electronic agreements, sent through Fiverr with screen names, were legally binding contracts. Exceeding the licensed scope was plausible infringement.

For businesses evaluating voice AI providers, that translates into a short checklist:

• Consented or licensed voices, with written terms covering specific uses, duration, and compensation
• Clear disclosure that callers are speaking with an AI, not a human
• Consent verification and usage monitoring, which law firms recommend alongside audio watermarking
• Jurisdiction-by-jurisdiction review, since state laws vary widely and the proposed NO FAKES Act would add federal watermarking and disclosure requirements

Disclosure deserves special attention. The NO FAKES Act, introduced in April 2025, would make watermarking and disclosure mandatory for digital voice replicas, and legal scholars argue that consent is required before training an AI on any individual's voice at all. Providers who treat these as design requirements — not afterthoughts — are the ones worth trusting.

Here's the practical takeaway: careful compliance practices are a marker of a trustworthy provider, in voice AI as in everything else. Any provider you consider — including Worqd — should be able to answer plainly how their AI voice systems handle consent, licensing scope, and caller disclosure. If the answer is vague, treat that as a signal.

When the foundation is right, the business case is real: an AI receptionist that answers every inquiry in under 60 seconds, books calls straight into your calendar, and hands off to a human with full context — without your business ever touching an unauthorized voice. Fast follow-up and legal soundness aren't in tension. Built properly, they come together.