What Does the Tennessee ELVIS Act Cover for AI Voice and Likeness?

What does the Tennessee ELVIS Act cover?

Governor Bill Lee signed House Bill 2091 in Nashville on 21 March 2024. The law, whose short title spells out ELVIS, updates Tennessee's right of publicity statute so that a person's voice is protected alongside their name, photograph and likeness. The enacted text defines voice as a sound in a medium that is readily identifiable and attributable to a particular individual, regardless of whether the sound contains the actual voice or a simulation of the voice. It took effect on 1 July 2024.

In practical terms, a song, advertisement or video that uses an AI clone of an identifiable person's voice without permission falls within the law's reach, as does a real recording used beyond what was authorised. Public reporting at the signing described it as the first state law aimed specifically at protecting musicians from AI voice imitation, and country artists attended the ceremony.

What changed from Tennessee's earlier right of publicity law?

Tennessee's Personal Rights Protection Act of 1984 was written for a world of photographs and endorsements. The ELVIS Act keeps its structure and extends it in two directions: to voice, including simulated voice, and to the technology used to imitate a specific person.

Who needs to pay attention to the ELVIS Act?

Record labels, publishers and streaming services are the obvious audience, because AI tracks imitating well-known singers were the trigger for the law. The scope is wider. Advertising agencies that use voice-over clones, game studios that synthesise character voices from actors, podcast and audiobook producers, and companies that sell voice-cloning services all handle identifiable voices.

The tool provision is what sets the law apart. A general text-to-speech engine has many lawful uses. A service built to produce one named person's voice on demand is closer to what the law targets. Voice AI companies with Tennessee users should be able to show what their product is for, how they obtain consent from the people whose voices they offer, and how they respond to complaints.

The law does not settle every question. How courts will apply its knowledge standards, exceptions and remedies to particular facts, such as tribute performances, interstate distribution or a model trained on many voices, has not yet been tested in published decisions.

What records should organizations keep for synthetic voices?

Most ELVIS Act disputes will turn on authorisation, so the best protection is a paper trail that answers it before anyone asks.

These records also protect the business that commissions synthetic work. If a performer later disputes a campaign, the company can show what was agreed and when, instead of relying on memory or email threads. Keep them for at least as long as the content stays in distribution.

• A signed licence naming whose voice or likeness is used, for which media, purposes, territories and duration.
• Whether the licence allows training a model on the voice, and whether the person can revoke it.
• The model, version and source recordings used to create each synthetic performance.
• An approval record before publication, and a complaint and takedown route afterwards.
• A list of distributors and platforms that received the content, so it can be withdrawn if rights lapse.

Where does deepfake detection help in a voice-likeness dispute?

A rights holder who suspects an unauthorised clone first needs to know whether a recording is synthetic at all. A forensic analysis that locates the audio segments most likely to be generated, and states its uncertainty, helps a label, artist or platform decide whether to send a takedown or a demand letter. Platforms that accept uploaded music or voice content can use the same analysis to route suspicious submissions to review.

Detection cannot answer the legal questions. It cannot tell whether the singer licensed the clone, whether an exception for news, commentary or parody applies, or what the publisher knew. A voice that sounds like a famous artist may also be a skilled human impersonator, which the law treats differently from a simulation built on the artist's recordings.

Good practice is to run detection on the best available copy, such as the uploaded master or the original file from the platform, rather than a re-recording captured from speakers.

How Neural Defend supports voice and likeness checks

NeuroDetect is Neural Defend's forensic workspace for authorised audio, video, image and document files. Each analysis in a case keeps its model and version context, detailed views go down to individual audio segments, and the downloadable forensic report records observations, uncertainty, provenance, model context and analyst notes. That format suits rights disputes, where the other side's expert will ask exactly how a conclusion was reached.

For likeness in images and video, NeuroVerify checks a face for manipulation and generative-model traces in one API call and returns signal-level evidence; on the public VendorBench-100 benchmark of adversarial still images, Neural Defend ranked first of 36 systems. For cloned voices in phone calls, NeuroCall is available in Beta. The ELVIS Act sits alongside federal rules such as the FCC's February 2024 ruling that AI-generated voices are artificial voices under the TCPA, so a single voice clone can raise both publicity-rights and telemarketing questions.