The Democratic midterm victory gives Congress an opportunity to revisit how artificial intelligence (AI) systems use music. Proposals address training data, copyright permissions and performers’ control over synthetic versions of their voices, though any legislation would face political and legal hurdles.
On November 5, the Nasdaq fell roughly 5.5% as investors weighed the prospect of tighter AI regulation and limits on data-center expansion. President Trump retains a veto, and bipartisan sponsorship of existing proposals does not guarantee their passage.
Copyright and commercial training
Copyright governs certain copying of compositions and recordings, subject to exceptions such as fair use. Whether particular AI training practices qualify for an exception remains disputed, and permission for a composition does not necessarily cover its recording.
In the 2025 Bartz v. Anthropic case, a judge found the training use at issue fair but rejected a fair-use defense for acquiring and retaining pirated books in a separate library. The decision did not resolve how copyright law applies to music training nationwide.
A discussion draft from Senator Marsha Blackburn proposes excluding unauthorized AI training from fair use. Another possible approach would require permission for defined commercial uses of protected music in generative AI training, with specified exceptions.
Records and transparency
Evidence of what went into a training dataset can be difficult for copyright owners to obtain. Similarity between a system’s output and an artist’s catalog alone does not establish that particular recordings were used or that infringement occurred.
- The CLEAR Act would require disclosures about covered copyrighted works in training datasets.
- The TRAIN Act would provide a subpoena process for owners seeking records concerning their own material.
Neither measure would automatically establish infringement. Record-retention rules and independent audits are further options for making training practices reviewable while keeping sensitive information available only to qualified investigators or a court. CLEAR’s coverage generally extends to registered works and certain older recordings, leaving out other music that can still be protected by copyright.
Voices, replicas and consent
Ownership of a master recording does not, by itself, settle whether a company may create a new performance using a singer’s voice. State protections vary, while the proposed NO FAKES Act would establish federal voice-and-likeness protections and a process for addressing unauthorized replicas.
Debate over the bill includes how to preserve room for reporting, criticism, parody and other protected expression. Protections for recognizable performers would also need to account for session singers, whose voices may have professional value without broad public recognition. A removal process could address replicas circulating before a lawsuit is resolved.
Voice replication also differs from musical influence: copyright generally does not grant ownership of a style, genre or production approach. Contract terms present a separate question of consent, including whether a session agreement or distributor upload authorizes later synthetic performances. California restricts certain digital-replica contract provisions, and SAG-AFTRA has sound-recording protections; the reach of those measures depends on their application.
