The ELVIS Act, Tennessee, and AI voice law

Tennessee was first. In March 2024, Governor Bill Lee signed the Ensuring Likeness, Voice, and Image Security Act into law. The acronym is not subtle. The state's economic identity is, in part, built on Elvis Presley's continuing commercial presence. The ELVIS Act extends Tennessee's existing right-of-publicity framework to cover voice specifically, and to explicitly include AI-generated likenesses.

What the Act does

The Act makes it unlawful to use a person's voice, name, photograph, or likeness without authorisation, including likenesses that are AI-generated or AI-cloned. It applies to commercial use. It applies to identifying voices even if the voice is technically generated rather than directly recorded.

Crucially, the Act extends to the estates of deceased Tennessee residents. The Presley estate, the Carter family, the entire Country Music Hall of Fame catalogue, every estate of a Tennessee-domiciled musician now has direct statutory protection against unlicensed AI voice work.

What the Act does not do

The Act is jurisdictionally limited. A voice AI vendor operating outside Tennessee, training on a deceased Tennessee resident's voice, and selling the result to a non-Tennessee buyer is in a grey zone. The Act is enforcement-dependent on a rights holder noticing and litigating. And it does not specify what consent or provenance documentation looks like in practice. The Act sets the prohibition; it does not define the cure.

This is the gap that careful methodology fills. Compliant practice requires documented consent, documented provenance, and a retirement protocol. The Act says these are necessary; the practitioner's job is to provide them.

What it means practically

The Act is the most important single piece of state legislation on this topic in the United States right now. Every estate trustee who manages a Tennessee-domiciled artist's catalogue should know about it. Every AI voice vendor who wants to work with these estates should be able to demonstrate documented consent, documented provenance, and a retirement protocol. Without those three, you are exposing your client to ELVIS Act liability and your own business to the lawsuit chain that follows.

Other states are following. California passed Assembly Bill 1836 six months later, with broadly similar effect for California-domiciled deceased performers. New York and Texas have similar bills in various stages of consideration. The state-level pressure is producing convergence; federal action is slower but moving in the same direction.

What good practice looks like

Five principles, in priority order.

Signed consent before any training begins. The rights holder, with documented authority from the governing trust or estate, signs an enumerated scope-of-use document. Anything not enumerated is forbidden.

Provenance documentation at every step. Source files cryptographically hashed. Preprocessing steps logged. Training runs logged with hyperparameters and software versions. Checkpoints hashed. Every deployment logged. Every inference call logged.

A retirement clause that is actually enforceable. The rights holder can retire the model at any time, on request, without justification. The retirement is verified at the infrastructure level. Self-hosted deployment is the only way to make this enforceable.

An ethical framework that holds together. The deceased is treated as material in the work, not as a simulated agent. No posthumous chatbots, no posthumous dialogue products, no voice clones that fake conversation. The operator authors the words; the model lends the voice.

No vendor lock-in. The client owns the trained model, the source code, the deployment, the documentation. They can run the system independently or retire it without us.

Four questions every buyer should ask

If you are an estate trustee, family executor, or institutional buyer considering AI voice work on your archive, the ELVIS Act now does most of the practitioner-vetting work for you. Ask any prospective vendor four questions.

Show me your consent and scope-of-use template. If they cannot, walk away.

Show me your provenance manifest format. If they shrug, walk away.

Show me your retirement protocol, including how deletion is verified. If they say "we delete files when asked," walk away. You want cryptographic attestation, not a promise.

Show me a case study with signed documents. If they have no case study, you are commissioning their first work, which is fine, but understand the implications and price accordingly.

Where this fits

Voice Estate, the methodology Alabo.Studio uses, is built around these principles. The methodology is published as a citable reference; it documents the six-phase process, the consent framework, the provenance manifest specification, the deployment patterns, and the retirement protocol. PERI, our flagship case study, is the worked example.

Other practitioners are building parallel approaches. Holly Herndon and Mat Dryhurst's Holly+ project established the consent-first playbook for living artists. Mozilla Common Voice demonstrated ethical corpus methodology. Fairly Trained, founded by Ed Newton-Rex, certifies training data practices. The Serpentine R&D Platform's Data Trust Experiment showed how institutional governance of voice data can work in practice. None of us are competitors; we are all building the same regulatory bridgehead.

The voice survives the artist. The ELVIS Act is the response to that technical capability. The practitioners who treat it as the foundation, not the constraint, are the ones who will do this work well.