Music – Part 2: Copyright & NIL

Introduction

Copyright law requires human authorship, but the threshold for how much human involvement qualifies remains undefined—prompts alone won’t get you there, and major labels are already rejecting recordings when AI use is disclosed. 

How artists can protect their voice and likeness from the ever-improving replicas generated by AI also remains an open question that is currently lacking a cohesive legal framework. There is no general federal right of publicity, and state laws vary significantly in coverage and remedies, though the pending federal legislation known as the NO FAKES Act aims to fill the gap. This section examines the existing legal protections, their limits, and the commercial considerations they raise.

Copyright and NIL 

The use of AI in music raises separate questions about copyright protection for the resulting AI-generated works and artists’ ability to control the replication of their voices and likenesses.

Copyright

U.S. copyright law requires human authorship. In Thaler v. Perlmutter, the D.C. Circuit affirmed the denial of registration for a work represented as autonomously generated by AI. The decision did not establish how much human involvement is sufficient to protect a work created with AI assistance.

For music, the question is which expressive elements a human authored. Using AI does not automatically disqualify an entire composition or recording from protection, but neither does directing an AI tool necessarily make its user the author of the resulting material.

In Part 2 of its Copyright and Artificial Intelligence report, the Copyright Office distinguishes assistive uses from uses in which AI determines the expressive content. It recognizes protection for human-authored expression that remains perceptible in an output, as well as sufficiently original human selection, coordination, arrangement, or modification of AI-generated material. Under the Copyright Office’s assessment of generally available technology, prompts alone do not provide sufficient control to establish authorship of the output. Protection extends to the qualifying human expression, not the purely AI-generated material itself.

The composition and sound recording also require separate consideration. Human-written lyrics or a human vocal performance may support protection even where other elements are generated by AI, but the scope of that protection will depend on the contributions to each work.

These distinctions may become increasingly difficult to apply as AI is integrated into songwriting and production tools. A finished track may reflect repeated exchanges between human and machine contributions, rather than neatly separable elements. Identifying which expression is human-authored and therefore protected can be difficult even with detailed records of the creative process. For artists, labels, and publishers, this creates uncertainty about the scope of the rights they can own, license, and enforce.

That uncertainty is already affecting commercial decisions. In our practice, we have seen major labels reject delivery of recordings after producers disclosed using AI in the production process. That caution appears to reflect uncertainty about when particular uses of AI affect copyright protection and how much of the finished recording remains protected.

Documentation can help establish the human contributions. In the registration of A Single Piece of American Cheese, an image created using AI-generated material, the Copyright Office recognized human authorship in the selection, coordination, and arrangement of that material. The applicant submitted a video documenting the creative process. As described in the applicant’s account, the basis for registration was human authorship, rather than prompting alone. Registration alone, however, does not conclusively establish the validity or scope of copyright protection.

For music creators, preserving session files, demos, stems, and revision histories can help show what they contributed. Registration applications should accurately identify the human authorship and, under the Copyright Office’s guidance, exclude more than de minimis AI-generated material from the claim. Those records can support a claim to protection, but they cannot create copyright in material that lacks human authorship.

Voice and likeness

There is currently no general federal right of publicity. Protection against unauthorized uses of an artist’s voice or likeness depends largely on the applicable state law, with significant differences in coverage, exceptions, and remedies. Most state publicity statutes predate generative AI and do not expressly address AI replicas, although existing protections may still apply. The Copyright Office identified these gaps in recommending federal legislation in its report on digital replicas.

These protections matter even when an AI-generated track does not infringe copyright. A simulated performance may be recognizable as that of an artist without copying protected expression from an existing composition or recording.

Tennessee’s ELVIS Act is among the newer state laws at the forefront of addressing AI replicas. It expressly protects identifiable voice simulations and imposes liability for knowingly distributing unauthorized uses of a person’s voice or likeness. Subject to a knowledge requirement, it also reaches providers of tools primarily designed to reproduce a particular person’s photograph, voice, or likeness. Qualifying labels and distributors have independent enforcement rights based on their exclusive recording-services or distribution agreements.

Some performers are pursuing trademark protection as another means of challenging AI imitations. Matthew McConaughey has obtained registrations covering specific audio and video clips, while Taylor Swift has reportedly applied to register sound clips and an image. This strategy remains largely untested. Registration does not confer ownership of a voice or appearance generally, and an infringement claim ordinarily requires a likelihood of confusion about source, affiliation, sponsorship, or approval. Federal false-endorsement claims may also be available without registration.

The NO FAKES Act seeks to establish federal protection directed at digital replicas. The Senate Judiciary Committee advanced S. 4591 in June 2026. We discussed an earlier proposal in AI and publicity rights: The No Fakes Act strikes a chord.

The committee version targets highly realistic, computer-generated representations readily identifiable as an individual’s voice or likeness. An exact clone is not necessarily required, but general stylistic resemblance alone would not suffice. The proposal includes expressive-use exceptions and notice, removal, and counter-notice procedures. Its coincidental-resemblance defense requires both purely coincidental resemblance and either a lack of public recognition or a lack of commercial value in the individual’s voice or likeness.

The proposed right would be nonassignable during life but licensable, subject to requirements governing form, duration, and permitted uses, with a collective-bargaining exception. Certain authorized uses could continue after license expiration or termination.

The bill would also let qualifying labels and distributors enforce the right based on exclusive recording-services contracts or exclusive licenses to distribute or transmit an artist’s recorded performances. Labels generally do not own or broadly control their artists’ publicity rights. This provision would give them independent enforcement authority without transferring those rights or allowing them to authorize replicas themselves.

For music transactions, agreements should separately address training permission, approval to simulate an artist’s voice, and exploitation of the resulting performance. They should specify who can approve each use, what compensation is available, and whether any uses may continue after the agreement ends.

Where things stand

• AI assistance does not automatically defeat copyright protection. Protection depends on human authorship. Preserving evidence of those contributions can support registration and enforcement, even as integrated AI tools make those contributions harder to separate.
• Voice and likeness protection remains uneven. State law remains central, while the NO FAKES proposal seeks a federal framework. Trademark claims offer a possible additional route, but their use against AI replicas remains largely untested.
• Enforcement authority and consent are separate questions. Giving a label authority to challenge unauthorized replicas does not itself give the label permission to create or license them. Agreements should clearly address approvals, permitted uses, and compensation.