Short answer, and the whole guide in one breath: it depends entirely on whose voice you clone. Cloning your own voice is generally fine, because it is your own likeness and you are consenting to it. Using a voice that is licensed or consented (a tool's own synthetic voice, or a real artist who agreed and is usually credited or revenue-shared) is generally fine too. Cloning a real, named artist without their consent is where it becomes legally and ethically fraught. The reason surprises most people: a voice on its own is not copyrightable, so the exposure is not mainly music copyright but the right of publicity (your control over your own likeness) plus a separate question about the copyrighted recordings a tool was trained on. Below we walk through the actual legal mechanisms in plain language, with verified sources, as of July 2026. This is general information, not legal advice; the law here is new and moving fast, and you should confirm a tool's current terms and consult a qualified attorney before you release anything.
Use a LICENSED or consented voice: generally fine. A tool's own synthetic voice, or a real artist who agreed (often credited or revenue-shared).
Clone a REAL named artist WITHOUT consent: legally and ethically fraught. Right-of-publicity exposure, takedowns, and in Tennessee criminal exposure under the ELVIS Act.
The mechanism: a voice itself is not copyrightable, so the main lever is the right of publicity (control of your likeness), plus a separate copyright question about the recordings a tool trained on.
This is general information, not legal advice. The law is new and changing; confirm current terms and consult a qualified attorney before releasing.
Why a voice is not copyrightable (and what that changes)
The instinct is to reach for copyright, but that is the wrong tool for a voice. Copyright protects fixed creative works: a specific sound recording, a composition, a set of lyrics. A person's vocal timbre, the raw quality that makes them sound like them, is not a fixed work and is generally not copyrightable in itself. That is why the legal action around AI voice cloning has centered on a different body of law: the right of publicity, which protects a person's control over the commercial use of their identity, including their voice and likeness. So when someone clones a named artist and releases it, the strongest claims are usually publicity and unfair-competition style claims about impersonating an identity, not "you copied my melody." There is a second, separate copyright question that does not concern your output at all: whether the AI model was trained on copyrighted recordings without a license. That one is being fought in court right now, and we cover it below.
Responsible use, in one line: clone only your own voice, or a voice whose owner has clearly consented (a tool's own synthetic voices, or a licensed artist roster). Never pass off a non-consensual clone of a real artist as that artist, and disclose that audio is AI-generated where required or where honesty calls for it. When in doubt, get written permission or pick a rights-first tool.
Which scenarios are safe, and which are not
The single most useful thing you can do is stop asking "is AI voice cloning legal" in the abstract and instead ask "whose voice, with whose permission, and am I releasing it." That reframes a fuzzy legal question into a checklist. Here is how the common scenarios line up, as of July 2026.
| Scenario | Legal status (general) | Why |
|---|---|---|
| Clone your own voice and release it | Generally fine | It is your own likeness and you consent; no other person's rights are implicated. |
| Use a tool's own synthetic voice (royalty-free) | Generally fine | No real, identifiable person's likeness is being used. |
| Use a licensed, consented artist voice | Generally fine | The artist agreed, usually with credit or a revenue share; permission is the whole point. |
| Clone a real named artist for a private experiment (not shared) | Lower risk, but check terms | No public commercial use, yet a tool's terms of service may still forbid it, and sharing changes everything. |
| Release a non-consensual clone of a real named artist | Legally and ethically fraught | Right-of-publicity claims, platform takedowns, and ELVIS Act criminal exposure in Tennessee. |
| Use a tool that trained on copyrighted recordings without a license | Unsettled, being litigated | The RIAA v Suno and Udio cases are testing exactly this; the law is not settled. |
The right of publicity: the main legal lever
Because a voice is not copyrightable, the primary legal exposure for cloning a real artist is the right of publicity, the right to control commercial use of your own identity. In the United States this is largely a matter of state law, so protection varies by state, but the direction of travel is unmistakable: lawmakers are extending it explicitly to AI voice cloning. The clearest example is Tennessee's ELVIS Act, the "Ensuring Likeness Voice and Image Security Act," signed in March 2024 and effective July 1, 2024. It is the first US law to protect a musician's voice alongside their name, image and likeness from unauthorized AI cloning, and a violation can be charged as a Class A misdemeanor, meaning cloning a Tennessee artist's voice without permission can carry criminal exposure, not just a civil lawsuit. If you clone and release a recognizable real artist without consent, this is the family of law that comes for the release.
A voice is not a copyright; it is an identity. The question is never "did I copy a song," it is "did I use a real person's likeness without their permission."The publicity-rights frame
The proposed federal answer: the NO FAKES Act
State-by-state publicity law is patchy, so there is a federal effort to standardize it. The NO FAKES Act would create a federal digital-replica right, giving individuals control over AI-generated replicas of their voice and likeness, with statutory damages reported in the range of $5,000 to $750,000 per violation. The crucial caveat for 2026: it is still a proposal. It has been introduced and reintroduced in Congress but, as of July 2026, has not been passed into law, so it does not yet bind anyone. Treat it as a strong signal of where federal law is heading, not as a rule you can rely on today. Anyone telling you the NO FAKES Act "makes AI cloning illegal" is getting ahead of the facts; for now the enforceable teeth are at the state level, with the ELVIS Act as the marquee example.
What the fake-Drake takedown actually showed
The case everyone remembers is "Heart on My Sleeve", the 2023 track by an anonymous creator using AI-cloned vocals of Drake and The Weeknd. It reached hundreds of thousands of streams, and Universal Music Group got it pulled from every major platform within roughly two weeks. It is cited as proof that AI voice cloning is illegal, but the nuance matters and cuts the other way. Reporting noted that UMG's cleanest, most unambiguous takedown hook was an unauthorized producer tag (a Metro Boomin sample) baked into the track, an actual piece of copyrighted material, rather than the cloned voices alone. In other words, the fastest legal lever was ordinary copyright on a sample, and the pure "you cloned our artist's voice" theory was, at that moment, still legally murkier. That murk is precisely why states began passing dedicated voice-and-likeness laws like the ELVIS Act. The takedown was real; the neat story that a voice clone alone was a slam-dunk copyright case is not.
The other lawsuit: training data, not your output
There is a second front that has nothing to do with what you release and everything to do with how the tool was built. In June 2024, the major record labels, through the RIAA, sued the AI music generators Suno and Udio, alleging they trained on copyrighted sound recordings without a license. This is a training-data copyright question, and it is unsettled and actively being litigated as of July 2026. The RIAA's stated position is worth understanding because it maps the industry's whole posture: licensed AI is embraced, unlicensed scraping is litigated. For you as a musician, the practical read is that tools built on clearly licensed or consented data carry less provenance risk, which is one reason certifications like Fairly Trained exist, to signal that a model's training data was properly licensed. It is a distinct issue from publicity rights, but it feeds the same conclusion: provenance matters.
Disclosure is becoming law: the EU AI Act
Even when your cloning is fully consented and clean, a newer obligation is arriving: telling listeners it is AI. Under Article 50 of the EU AI Act, AI-generated audio must be disclosed and marked in a machine-readable way, with these transparency obligations coming into force on August 2, 2026. If you distribute into the EU, this is a concrete, dated requirement rather than a nice-to-have. Beyond the law, disclosure is simply the ethical default that keeps AI vocals from being deception, and it is why rights-first tools increasingly build watermarking and labeling in. Note that popular streaming platforms are also evolving their own AI-content policies; those change frequently, so we will not quote a specific platform rule here, only flag that you should check the current policy of wherever you distribute.
Responsible use, made concrete
Put together, the safe path is not complicated, and it lets you use these tools with a clear conscience and a clean release. The whole discipline collapses to consent and provenance.
The responsible-use checklist: (1) clone only your own voice, or use a tool's own synthetic or licensed, consented voices; (2) get written permission before you clone any real person, every time; (3) prefer tools with clean provenance, such as Fairly Trained members; (4) disclose that audio is AI-generated where required (EU AI Act) or where honesty demands it; (5) never present a non-consensual clone as the real artist; (6) confirm the tool's current terms and consult a qualified attorney before a commercial release.
The law is moving: a quick timeline
This field went from lawless to legislated in about three years, which is why last year's advice is already stale. Here is the sequence that shaped where things stand in July 2026.
Frequently asked questions
Is it legal to clone my own voice with AI? Generally yes. It is your own likeness and you are consenting, so the right-of-publicity concern that applies to cloning other people does not apply to you. Confirm the tool's terms and, if you are releasing commercially, that the output is licensed for that use. See can you release AI vocals commercially?
Is it illegal to clone a famous singer's voice? Releasing a non-consensual clone of a real, named artist is legally and ethically fraught. It can trigger right-of-publicity claims and platform takedowns, and in Tennessee it can carry criminal exposure under the ELVIS Act. A private experiment you never share is lower risk, but a tool's terms may still forbid it. Get written consent, or do not do it.
Can I copyright someone's voice, or is it copyrighted? A voice on its own is generally not copyrightable; it is protected instead through the right of publicity (control of your likeness). A specific recording of that voice can be copyrighted, but the raw vocal quality is not.
Did the NO FAKES Act make AI voice cloning illegal? No. As of July 2026 the NO FAKES Act is still a proposal that has not passed into law. The enforceable protections today are state laws like the ELVIS Act plus existing right-of-publicity doctrine.
Do I have to disclose that vocals are AI-generated? Increasingly, yes. Article 50 of the EU AI Act requires disclosing and machine-readable-marking AI-generated audio, with obligations in force from August 2, 2026, and disclosure is the ethical default everywhere. Streaming platforms also set their own evolving policies, so check the current rules where you distribute.
Which tools are the safest bet for staying on the right side of this? Tools built on consented, licensed voices with clear provenance, several of which carry Fairly Trained certification. Our contrast of a rights-first tool against a clone-anything one is in Kits AI vs Musicfy, and the full field is in the best AI vocal tools roundup.
Bottom line
Is AI voice cloning legal? Ask a sharper question: whose voice, and did they consent. Cloning your own voice or using a licensed, consented voice is generally fine; cloning a real named artist without consent is where you run into the right of publicity, dedicated laws like the ELVIS Act, and platform takedowns. A voice is not copyrightable, so this is about likeness and permission, with a separate, still-unsettled fight over the copyrighted data these tools were trained on. The law is being written in real time (the NO FAKES Act is proposed, EU disclosure lands in August 2026), so the durable move is to build on consent and provenance: clone only your own or a consented voice, prefer clean-provenance tools, disclose when appropriate, confirm current terms, and get a real lawyer for a real release. This is general information, not legal advice, as of July 2026.