Back to Law Hub
Guides & Knowledge

AI Voice Cloning in Music: The Legal Risk Beyond Copyright

Legal Team11 min read|

An AI can clone a singer's voice without copying a single recording, which is why the legal risk of AI voice cloning in music often lies beyond copyright.

Disclaimer: This article is for general informational purposes only and does not replace legal advice for specific situations.

TL;DR

Under U.S. law, a voice is generally not copyrightable: copyright protects the fixed sound recording and the underlying composition, not the vocal identity an AI clone imitates [1][2]. Protection against voice cloning therefore comes from outside copyright — state right-of-publicity law, voice-specific statutes, contract terms, and platform policies — while a federal voice right remains a pending bill. Outside the U.S., regulators are moving along two separate tracks: consent rights and mandatory labeling. For creators and licensees, the core risk lies in assuming that clearing the copyright also clears the voice.

Key definitions: Three layers inside one AI vocal

Every recorded vocal typically carries at least three distinct legal layers, and an AI voice clone interacts with each one differently. Copyright attaches to the first two once they are fixed in a tangible medium; the third — the voice as an attribute of a person's identity — sits outside copyright [1][2].

LayerWhat it coversTypical rights holderGoverning regimeDoes an AI clone of the voice infringe it?
Musical CompositionMelody, harmony, lyricsSongwriter / music publisherCopyright (composition)Only if the AI output reproduces protected melody or lyrics — imitating the voice alone typically does not
Sound RecordingThe specific fixed audio performanceRecord producer/ Record labelCopyright (sound recording)Only if the actual recorded sounds are duplicated or altered; a new recording that merely imitates them generally does not infringe [2]
Voice / vocal identityThe recognizable sound of a specific personThe individual (or their estate)Right of publicity, voice statutes (not copyright). Permission to use a voice is usually granted by contract, but a contract binds only the parties who sign it.Potentially yes, but under non-copyright law, and only where that regime's conditions (e.g., identifiability, commercial use) are met [1][3]

The distinction has practical weight because composition and recording rights are often held by publishers and labels, while the voice remains personal to the singer. A license from a publisher or label covers the layers those parties control; it does not, on its own terms, extend to the performer's vocal identity.

See more: "Which parts of a song are protected by Copyright?"

In the United States, an AI voice clone is assessed under two bodies of law that work differently. Federal copyright law governs the composition and the sound recording but does not protect the voice itself, so it largely defines what a voice clone does not infringe [1][2]. Protection for the voice as part of a person’s identity comes mainly from state right-of-publicity laws, which vary in whether, and how far, they extend to voice, imitation, and digital replicas [1]. Other federal laws, such as the Lanham Act’s false endorsement provision, reach voice imitation only in narrow circumstances [1].

Framework 1: Copyright protects the recording, not the voice

Section 114(b) of U.S. Copyright Law limits the exclusive rights of sound recording owner to the actual sounds fixed in theirs recording. Accordingly, an independently created recording that merely imitates those sounds does not, by itself, constitute copyright infringement[2]. In most cases, an AI-generated clone of a singer's voice falls into this category.

This distinction arises because copyright protects original expression embodied in a work, rather than the underlying characteristics or qualities of a person’s voice. Therefore, a traditional infringement claim generally fails unless the clone also copies protected melody, lyrics, or recorded sounds. The U.S. Copyright Office confirmed in 2024 that a voice, as such, is not copyrightable [1]. Similarly, in Lehrman v. Lovo, Inc. (S.D.N.Y., 2025), the court applied the same rule in addressing claims involving AI-generated voice clones [4].

Framework 2: Voice protection under state right-of-publicity laws

The United States has no federal right of publicity; Instead, protection against the unauthorized commercial exploitation of an individual's identity, including name, likeness, and voice is governed by state law [1]. Courts have protected voices on that basis. In Midler v. Ford Motor Co., the Ninth Circuit held that deliberately imitating a well-known singer’s voice to sell a product is a tort under California law [3], and in Lehrman v. Lovo the voice actors’ claims under New York’s right-of-publicity statute survived [4].

State lawVoice protectionStatus
Tennessee ELVIS ActCovers actual and simulated voice; liability extends to distributors of tools whose primary purpose is producing an identifiable person’s voice without authorization [5]In force since July 1, 2024
New York Civil Rights Law §§ 50–51Prohibits using a person’s voice for advertising or trade without written consent [4]In force
California AB 2602 / AB 1836Makes digital replica clauses unenforceable unless intended uses are described; requires estate consent for replicas of deceased performers, including in sound recordings [6]Signed September 2024

At the federal level, the NO FAKES Act (S. 4591) would create a licensable right in voice and likeness against unauthorized digital replicas [14]. The Senate Judiciary Committee advanced it on June 18, 2026 [7], and it was placed on the Senate Legislative Calendar on June 24, 2026, with no floor vote recorded [14]. Platform rules add a further layer: Spotify’s policy, for example, permits vocal impersonation only where the artist has authorized it [8].

See more: "AI-Generated Music Sounds Like My Style — Is That Copyright Infringement?"

Outside the U.S., regulation is developing along two distinct tracks that answer different questions. Consent-based rules ask whether a voice clone may be published at all; labeling rules ask whether audiences must be told the content is synthetic.

JurisdictionInstrumentMechanismStatus (October 2026)Requires the voice owner's consent?
European UnionAI Act, Article 50 [9]Deployers must disclose deepfake audio, image, and video contentApplies since August 2, 2026; machine-readable marking for generative systems already on the market extended to December 2, 2026 [10]No — disclosure only; a labeled clone may still be unlawful under other law [9]
VietnamLaw on Artificial Intelligence No. 134/2025/QH15 [13]Deployers must clearly label AI-generated audio that simulates a real person’s voice; providers must apply machine-readable marking.In force since March 1, 2026 [13]No — the obligation addressed is labeling by deployers.
DenmarkBill L 49, amending the Danish Copyright Act [11][12]Prohibits making a realistic, digitally generated imitation of a person’s voice or appearance available to the public without that person’s consent; also covers digital imitations of performers’ performances [12]Tabled in the Folketing on October 8, 2026; targeted to take effect January 1, 2027; not yet enacted [12]Yes, if enacted, subject to exceptions for satire, parody, and social or political criticism [12]

In the EU, the Digital Omnibus package adopted in June 2026 postponed several AI Act obligations but left the deepfake disclosure duty on its original timeline; the only Article 50 extension concerns machine-readable marking for generative systems already on the market [10]. The disclosure duty is a transparency rule — it does not make an unauthorized clone lawful [9].

Denmark's proposal is the clearest attempt to give individuals a consent right over their voice, yet even it sits only nominally inside copyright. EU research analysis notes that the Danish explanatory materials do not intend to create a new copyright for citizens, and that several Member States consider deepfake harms to extend beyond the remit of copyright law [11]. The bill would also cover digital imitations of performers' performances [12] — an area traditionally associated with performers' neighboring rights.

See more: "What are Neighboring Rights?"

Vietnam's Law on AI addresses AI-generated audio simulating a real person's voice through a labeling obligation on AI system suppliers [13]. For creators releasing music in Vietnam, this means a synthetic vocal may need to be identified as such — a separate question from whether the person whose voice is simulated agreed to it.

Common misconceptions

MisconceptionWhat the law actually saysWhy it matters
"I own my master recordings, so my voice is copyrighted."Sound recording rights cover the actual fixed sounds; independent imitations fall outside them, and a voice itself is generally not copyrightable [2][3]A copyright claim against a pure voice clone may fail; the viable claim often sits in publicity or contract law [4]
"If the track is labeled 'AI-generated', releasing it is lawful."EU and Vietnamese labeling rules concern transparency, not consent [9][13]A labeled clone may still breach publicity rights, platform policy likes Spotify, or — if enacted — Denmark's L 49 [8][12]
"An AI voice tool's 'commercial license' covers any use."In Lehrman v. Lovo, the court allowed consumer protection claims premised on a provider allegedly overstating the commercial rights its voice clones carried [4]A licensee may face exposure if the tool provider never obtained the voice owner's consent

Practical takeaway

Three points typically determine who carries the risk when an AI vocal resembles a real singer:

  • Copyright clearance is not voice clearance. Licensing a composition from a publisher and a master from a label covers those layers; permission to synthesize the performer's voice generally has to come from the individual or their estate [1][6].
  • Liability can extend beyond the uploader. AI providers have faced publicity and consumer protection claims [4], Tennessee's statute reaches parties beyond the direct user [5], and the pending federal bill would expose platforms that knowingly host unauthorized replicas [7].
  • Consent terms carry the weight. California already treats contract provisions on digital replicas as requiring a description of intended uses [6]. Consent drafted too broadly may be challenged as unenforceable; consent drafted too narrowly may not cover the actual release format, territory, or platform.

Conclusion

AI voice cloning raises a question that copyright law largely leaves open. Copyright protects the composition and the sound recording, while the voice itself is protected, where it is protected at all, by right-of-publicity law, voice-specific statutes, contract terms, and platform rules that vary by jurisdiction. For creators and licensees, the practical risk rarely comes from deliberate copying. It more often comes from treating a copyright license as if it also covered the performer's voice, or from relying on an AI tool's terms without confirming that the person whose voice was modeled agreed to that use.

The legal question is no longer whether a voice can be protected, but under which law: since August 2, 2026, synthetic audio released in the EU carries disclosure duties [9], Denmark's consent rule is targeted for January 2027 [12], and in the United States the answer still turns on state lines.


REFERENCES

[1] U.S. Copyright Office, Copyright and Artificial Intelligence, Part 1: Digital Replicas (July 2024);

[2] Section 114(b), Title 17, U.S. Code (17 U.S.C. § 114(b)), Copyright Act of 1976;

[3] Midler v. Ford Motor Co., 849 F.2d 460 (9th Cir. 1988);

[4] Lehrman v. Lovo, Inc. (S.D.N.Y. July 10, 2025); New York Civil Rights Law §§ 50–51;

[5] Tenn. Code Ann. § 47-25-1101 et seq., Ensuring Likeness, Voice, and Image Security (ELVIS) Act (2024);

[6] Cal. Labor Code § 927 (AB 2602) and Cal. Civil Code § 3344.1 (AB 1836) (2024);

[7] Office of Sen. Marsha Blackburn, "Blackburn, Coons Bipartisan Bill to Protect Individuals and Creators from Deepfakes Passes Senate Judiciary Committee" (June 18, 2026);

[8] Spotify Newsroom, "Spotify Strengthens AI Protections for Artists, Songwriters, and Producers" (September 25, 2025);

[9] Article 50, Regulation (EU) 2024/1689 (Artificial Intelligence Act) (2024);

[10] Ropes & Gray, "Operationalising the EU AI Act's Transparency Obligations" (August 2026);

[11] European Parliamentary Research Service, At a Glance ATA(2026)782611, Denmark's Deepfake Proposal (January 2026);

[12] Euronews, "Denmark Introduces Bill to Protect People's Faces and Voices from Deepfakes" (October 8, 2026);

[13] Law on Artificial Intelligence No. 134/2025/QH15 of Vietnam (2025).