Is AI Voice Cloning Legal in the US in 2026? A Creator's Safe-Use Guide

#What we're covering

1. TL;DR — the four laws you have to know
2. Federal law: TAKE IT DOWN Act & the FTC impersonation rule
3. Tennessee's ELVIS Act — the first state to make cloning a crime
4. California AB 2602 and AB 1836 — voice as intellectual property
5. Other states with active AI-voice laws
6. What ElevenLabs actually enforces (Voice Captcha, consent, moderation)
7. Safe-use rules for podcasters, YouTubers, and ad agencies
8. The pre-publish checklist
9. FAQ

#TL;DR — the four laws you have to know

If you skip the rest of this article, memorize this list. Every one of these is currently in force in the US.

Notice what these laws do not do. None of them ban AI voice cloning outright. They ban it when it happens without consent, when it is used to defraud, or when it produces sexual deepfakes. Legitimate cloning — your own voice, a voice actor who signed a proper release, a synthetic voice with no real-person basis — is still perfectly legal. The trick is knowing which category your specific use falls into.

#Federal law: TAKE IT DOWN Act and the FTC impersonation rule

TAKE IT DOWN Act (signed May 2025, enforceable in 2026)

The full name is the "Tools to Address Known Exploitation by Immobilizing Technological Deepfakes on Websites and Networks Act," which is a Congress acronym at its finest. It was signed by President Trump on May 19, 2025, and became fully enforceable on the standard one-year clock, meaning it is active law in 2026.

Two things it does that matter for AI voice users:

1. Criminal ban on non-consensual intimate imagery — including AI-generated content that depicts a real person in a sexual context without their consent. This covers "morphed" content: think a real voice paired with sexual audio or dialogue that person never actually said. The offender faces up to three years in federal prison.
2. Platform takedown mandate — covered online services (basically any site that hosts user content) must remove non-consensual intimate deepfakes within 48 hours of a valid notice from the victim. If you are a creator running a YouTube channel or a podcast platform, you are almost certainly not the target — but if your AI voice content gets flagged, the platform has 48 hours to act, and they will act first and ask questions later.

The practical takeaway: if you are generating anything with a real person's voice that could be construed as sexual or intimately embarrassing, you are firmly in federal-crime territory. This includes "parody" or "prank" content where you use a celebrity or ex-partner's voice in a compromising context. The TAKE IT DOWN Act does not care about your comedic intent.

FTC Government and Business Impersonation Rule (2024)

The Federal Trade Commission finalized this rule in April 2024, and it specifically covers AI voice cloning used to impersonate a government agency or business. This is the rule that got a lot of coverage when scammers used AI voice of Joe Biden to discourage voting in New Hampshire — that call operator, Steve Kramer, was charged and fined heavily under related laws.

What triggers it:

• Using an AI voice clone that sounds like a real government official (IRS, Social Security, local police) in a fraudulent context
• Cloning a business's voice or spokesperson to trick consumers into paying, sharing data, or clicking links
• The commercial-fraud angle is what makes it FTC territory rather than pure First Amendment territory

Penalties are civil, not criminal, but they are steep — the FTC can seek over $50,000 per violation, and "per violation" often means per call or per instance. A scam campaign with a thousand calls is a fifty-million-dollar exposure. This mostly applies to bad actors, but if you are producing "prank calls" content for social media using AI voice, you have to be very careful about the impersonation-of-a-business angle.

#Tennessee's ELVIS Act — the first state to make voice cloning a crime

Tennessee passed the Ensuring Likeness Voice and Image Security Act (ELVIS Act) in March 2024, and it went into effect on July 1, 2024. Musicians pushed hard for this one — the state houses the music industry, and Nashville had been watching AI-cloned Drake tracks go viral for a year. The Recording Academy, the RIAA, and independent artists all lobbied for it.

What makes ELVIS distinct from most state right-of-publicity laws:

• Voice is treated as a standalone property right — before ELVIS, most states protected "name, image, and likeness" and left voice as a fuzzy afterthought. ELVIS gives voice the same standing as your face.
• It applies to anyone whose voice was cloned, not just celebrities. If someone clones your voice in Tennessee without your written consent, you have a cause of action.
• Criminal penalties — up to a Class A misdemeanor, meaning up to a year in county jail and fines up to $2,500. Plus civil damages on top.
• It covers "algorithms, software, tools, or other technologies, services, or devices" — meaning it targets not just the end user but the platforms that knowingly facilitate unauthorized cloning.

Real-world impact for creators: if you clone anyone's voice without a written release, and the person or the content ends up in Tennessee, you are exposed. This does not mean you need to avoid Tennessee — it means you need consent. A signed release form is your friend.

#California AB 2602 and AB 1836 — voice as intellectual property

California signed two related bills in September 2024, both effective January 1, 2025. Together they create the most sophisticated voice-cloning framework in the US.

AB 2602 — Living performers

This one is aimed at the entertainment industry and it explicitly targets contract language. If a studio, agency, or client wants to use a digital replica of a performer's voice (or face, or body), the contract has to:

• Include a "reasonably specific" description of how the digital replica will be used
• Be signed by the performer or a labor union representing them
• Be entered into with the performer represented by legal counsel or a union rep, or the contract is unenforceable

Practical translation: if you hire a voice actor and want to clone their voice for future use, you cannot get away with a vague "all rights" clause. You have to say exactly what you will use it for. "For any purpose in perpetuity" is now unenforceable if the actor didn't have union or lawyer representation at signing.

AB 1836 — Deceased performers

Extends the same digital-replica protections to the estates of deceased personalities. If you want to bring back James Dean or Whitney Houston's voice, you need permission from their estate, and the same specificity requirements apply. Statutory damages start at $10,000 per instance plus profits, or the estate's actual damages, whichever is greater.

This is the law that shut down some of the "AI-resurrect a dead celebrity" novelty apps that were floating around in 2024. It is now genuinely risky to use a deceased celebrity's voice in commercial content without an estate license.

#Other states with active AI-voice laws

Beyond Tennessee and California, several states have added meaningful voice-cloning provisions. This is not exhaustive — the map is changing every legislative session — but as of mid-2026:

• New York — Extended right-of-publicity to cover digital replicas of deceased performers in 2020, and has ongoing legislation to cover the living. Voice is explicitly included.
• Illinois — The Right of Publicity Act was amended to explicitly cover AI-generated digital replicas. Statutory damages available.
• Louisiana — Passed the Allen Toussaint Legacy Act (2022, updated 2024) covering voice as protected likeness.
• Utah — 2024 amendments to publicity rights explicitly cover synthetic media.
• Washington — Has ongoing efforts on political-deepfake laws that overlap with voice cloning.

The pattern is clear: states with strong media, tech, or entertainment industries are moving fastest, and voice is being carved out as its own protected right. If your audience is nationwide (and if you are on YouTube or a podcast platform, it is), you have to assume the strictest applicable state law governs your content.

#What ElevenLabs actually enforces

ElevenLabs is the market leader for AI voice cloning, so a lot of creators default to it. It is worth knowing what the platform itself does to keep you (and them) out of legal trouble. As of 2026 they have four main mechanisms:

1. Voice Captcha for Instant Voice Cloning (IVC)

When you upload a short audio clip to instantly clone a voice, ElevenLabs will sometimes require you to re-record a randomly generated verification phrase in the same voice, live. This is designed to make sure you are cloning your own voice, not someone else's from a downloaded YouTube clip. It is not foolproof, but it introduces meaningful friction for bad actors.

2. Voice Captcha for Professional Voice Cloning (PVC)

The higher-quality Professional Voice Cloning tier requires 30+ minutes of clean audio and a legally binding attestation that you either own the voice or have written permission. Skipping the attestation is fraud on the platform, and ElevenLabs has terminated accounts for it.

3. Moderation on politically sensitive prompts

ElevenLabs will refuse to generate content that names US political figures, election-related speech, or certain other categories tied to fraud risk. You can get around this with jailbreaks, but you are then in violation of their ToS and directly liable if the output causes harm.

4. AI Speech Classifier

ElevenLabs offers a free tool that lets anyone (victim, platform, journalist) check whether an audio clip was likely generated by their platform. This creates traceability that other platforms lack. If your synthetic content gets flagged, it becomes part of a chain of custody that can lead to legal exposure.

Similar mechanisms exist at PlayHT, Resemble AI, Speechify, and Descript's Overdub — the specifics differ, but the pattern of consent attestation plus moderation is now industry standard for the major players. The smaller and newer platforms are where corners get cut, and where your legal exposure is highest.

#Safe-use rules for podcasters, YouTubers, and ad agencies

Here is the practical part. If you are a working creator, these are the categories of AI voice use ranked from bulletproof to actively-illegal.

Green zone — legally safe (do these freely)

• Clone your own voice for a podcast intro, ad-read insertion, or "second host" persona. You cannot infringe your own rights. Save the attestation email from your platform as a paper trail.
• Use fully synthetic voices with no real-person basis — ElevenLabs' "Alex," "Nicole," "Rachel," etc., that are trained on paid voice actors under permissive contracts. These are meant to be used royalty-free for commercial content.
• Clone a voice actor's voice with a properly executed consent form that spells out the exact usage (podcast, YouTube, ad, in what markets, for how long). This is the AB 2602 pattern.
• Text-to-speech for accessibility — screen-reader-style narration of your own written articles or documents. No cloning involved, no legal issue.

Yellow zone — legal but risky (get counsel first)

• Parody using a real public figure's voice, especially politicians. First Amendment protects true parody, but the line between "parody" and "misinformation" is drawn by prosecutors and juries, not by you. If the content is at all plausible, you are on thin ice.
• Voice cloning a deceased celebrity for tribute content. Unless you have estate permission, both AB 1836 (California) and similar state laws create serious exposure.
• Cloning a voice for internal-use-only training material that then leaks. Consent for internal use is not consent for public distribution. Wrap it in NDAs.

Red zone — do not do these

• Anything sexual involving a real person's voice without their explicit written consent. Federal crime under TAKE IT DOWN.
• Impersonating a government agency, business, or bank in any commercial or defrauding context. FTC territory, and every state also has consumer-protection laws that pile on.
• Cloning a specific person to make them "say" something they never said for the purpose of misleading viewers, even if the content is not sexual. This is defamation-per-se in most states.
• Political disinformation — an AI-cloned voice of a candidate distributed in the 60 days before an election is now specifically covered by laws in over a dozen states.

#The pre-publish checklist

Every time you are about to publish AI-voice content — a podcast episode, a YouTube video, a client ad — run through this list. If any answer is unclear, stop and get advice.

1. Is the voice mine, fully synthetic, or a voice actor's with a signed release? If none of those, do not publish.
2. Does the content involve sexual, intimate, or embarrassing subject matter tied to a real person? If yes, absolute stop.
3. Does the content impersonate a government agency, bank, or business in a way that could deceive consumers? If yes, stop.
4. Am I using a deceased celebrity's voice? If yes, do you have estate permission in writing?
5. Have I disclosed AI voice usage to my audience? Not legally required in most cases, but strongly recommended — it builds trust and pre-empts complaints to platforms.
6. Do I have a paper trail? Save consent forms, ElevenLabs attestations, and platform receipts for at least seven years. Statutes of limitation are long.
7. If I am running ads with this voice, does the client know and approve? Ad agencies are increasingly requiring AI-voice disclosure in their creator contracts. Get it in writing.

If you want to sanity-check your workflow against a real free voice tool, our own AI Voice Generator uses fully synthetic voices (Cloudflare Workers AI, Kokoro TTS, Qwen3, and Web Speech), so there is no cloned-voice legal exposure for typical narration or explainer use. It is a good starting point if you want to test scripts before committing to a paid ElevenLabs plan for a real client project.

#FAQ

#Bottom line

AI voice cloning is not going away, and the good news is that using it responsibly is not hard. Clone your own voice. Use fully synthetic voices when you need variety. Get proper releases from voice actors. Never touch anyone else's voice without written consent. Save your paperwork. That is 90 percent of the safe-use protocol.

The remaining 10 percent is the judgment calls — parody, tribute content, political commentary — where you should get actual legal advice before you publish. It is worth the hour of a media attorney's time to keep your business alive. The four laws above have real teeth, and the enforcement environment is only getting stricter through 2026.

For a free, no-legal-risk starting point that uses only synthetic voices, try our AI Voice Generator. And if you want to compare the major voice-cloning platforms before you commit to one, our best AI voice generators comparison has the current 2026 lineup with pricing, quality, and consent-workflow notes for each.