The United States Copyright Office (USCO) has released the first in a series of reports relating to the implications of AI on copyright law and policy.  This initial report, titled “Copyright and Artificial Intelligence Part 1: Digital Replicas,” focuses on AI-generated digital replicas commonly known as “deepfakes,” and which utilize digital technology to replicate a person’s appearance or voice. 

The report comes out of the Copyright Office’s AI Initiative, launched in early 2023 to examine the copyright issues raised by AI, and follows public listening series, webinars, and a notice of inquiry which garnered over 10,000 comments from the public.  Surveying the state of existing law and concluding that “there is an urgent need for a robust nationwide remedy beyond those that already exist,” the report recommends parameters of proposed new federal legislation addressing deepfakes, to include:

Focus on highly realistic replicas – The report emphasizes the need for a statute that specifically targets digital replicas that are so realistic they are indistinguishable from genuine depictions (whether or not generated by AI). USCO suggests that this protection be narrower and distinct from the broader “name, image, and likeness” protections currently offered.

All persons protected – USCO recommends extending protections to all individuals, not just celebrities or those with commercially valuable identities, representing an expansion of many state right of publicity laws.

Extendable lifetime protection – The report suggests that statutory protections last for an individual’s lifetime with any postmortem protection limited in duration, but extendable if the individual’s persona continues to be exploited after death.

Carve out for creation – USCO urges that liability should arise from distributing or making available unauthorized digital replicas, but not from creation alone. The Office also proposes that liability extend beyond commercial uses, but that actual knowledge that a digital replica was unauthorized be required for its imposition.

Secondary liability and safe harbor – The report recommends that copyright law’s concepts of contributory liability and vicarious liability should apply, with a safe harbor similar to that under the DMCA to encourage online service providers to promptly remove deepfakes.

Licensing permitted, assignments not – USCO suggests that individuals be able to license and monetize their digital replica rights but not be allowed to assign them outright. Licenses involving minors would have additional protections.

Avoiding First Amendment conflicts – USCO urges that the statute expressly address free speech concerns using a balancing framework (rather than categorical exclusions) to avoid overbreadth and allow greater flexibility.

Injunctive relief and money damages recommended – The report proposes that the law provide remedies in the form of injunctive relief and monetary damages. USCO also notes that criminal liability might be appropriate in some cases.

No preemption – USCO recommends that the law not fully preempt state law, including state right of publicity and privacy law. USCO envisions that the federal law would provide a baseline of protection nationwide, leaving states with the power to provide additional protections.

This report comes at a time when deepfakes are top of mind for lawmakers.  Just prior to the report’s publication, the U.S. Senate passed the DEFIANCE Act to provide recourse for victims of sexual deepfakes.  And the same week as its issuance, the NO FAKES Act was introduced in the Senate with the goal of providing broader legal recourse against deepfakes, more generally.  Additionally, and with this publication, the US Patent and Trademark Office now has until January 27, 2025 (under President Biden’s AI Executive Order), to issue its own recommendations for executive action on copyright and AI.

Forthcoming reports in the Copyright Office’s series are expected to cover the copyrightability of AI-generated works, training of AI models with copyrighted material and licensing.  Updates to the Compendium with respect to the registrability of AI generated works are also anticipated.

Authored by Lauren Cury and Ryan Thompson