The United States Copyright Office (USCO), a unit of the Library of Congress, issued a report in 2024 containing recommendations on the regulation of deepfakes - synthetic media created with artificial intelligence that can depict real people saying or doing things they never did. The report, titled "Copyright and Artificial Intelligence, Part 3: Digital Replicas," was the third in a series examining copyright issues raised by generative AI, and it proposed a new federal law to protect individuals from unauthorized digital replicas of their voice or likeness.
The Copyright Office's involvement in deepfake policy stems from its broader mandate to advise Congress on copyright matters. As part of the legislative branch, the Office assists in developing national and international copyright policy, drafting legislation, and preparing technical studies. The deepfake report was a direct response to requests from the Senate Judiciary Committee's Subcommittee on Intellectual Property and the House Judiciary Committee's Courts, Intellectual Property, and the Internet Subcommittee, which asked the Office to analyze the copyright implications of AI.
Background and Legal Context
The Copyright Office has long administered the U.S. copyright system, which traces its origins to the Copyright Act of 1790. That first federal copyright law covered only books, maps, and charts, with claims recorded by clerks of U.S. district courts. The first registered work was the Philadelphia Spelling Book by John Barry, filed on June 9, 1790, with the U.S. District Court of Pennsylvania. In 1870, copyright functions were centralized in the Library of Congress under Librarian Ainsworth Rand Spofford, and the Copyright Office became a separate department on February 19, 1897, with Thorvald Solberg as the first register of copyrights.
The Copyright Act of 1976, signed by President Gerald R. Ford on October 19, 1976, and effective January 1, 1978, made registration largely optional for copyright ownership, requiring only fixation of an original work in a tangible medium. However, registration (or refusal) remains a prerequisite for infringement lawsuits. The Office maintains the Copyright Card Catalog, containing nearly 45 million individual cards indexing registrations from 1870 to 1977, and an online database with more than 16 million entries for records after 1977.
Deepfakes implicate several areas of existing law, including copyright, right of publicity, and defamation. The Copyright Office's report focused on the copyright and related intellectual property aspects, particularly the concept of a "digital replica" - a sound recording or image that is created with AI to resemble a specific individual's voice or appearance. The Office noted that existing state laws on right of publicity vary widely, creating a patchwork of protections that may not adequately address the national and international scale of deepfake distribution.
The 2024 Report on Digital Replicas
The Copyright Office released its report on digital replicas on July 31, 2024. The report concluded that current federal law does not provide sufficient protection against the harms caused by unauthorized digital replicas, and it recommended that Congress enact a new federal statute to address this gap. The proposed law would create a civil action for individuals whose voice or likeness is used in a digital replica without authorization, with liability for those who knowingly distribute or make available such replicas.
The report also recommended that the new law include a safe harbor for online service providers, similar to the Digital Millennium Copyright Act (DMCA) notice-and-takedown system, to balance protection with the free flow of information. The Office suggested that the law should not preempt state right-of-publicity laws but rather establish a federal floor of protection. It also addressed the First Amendment concerns, recommending that the statute include exemptions for news reporting, commentary, criticism, and other expressive uses, while allowing for post-hoc liability in cases of actual harm.
The report was part of a broader series on AI and copyright. Part 1, released in 2023, addressed the use of copyrighted works in training AI models. Part 2, released in early 2024, examined the copyrightability of AI-generated outputs. Part 3 focused specifically on digital replicas, and Part 4, released later, covered other topics such as collective licensing.
The Register of Copyrights and Leadership
The report was issued under the leadership of Shira Perlmutter, who became the 14th Register of Copyrights on October 26, 2020. Perlmutter, a former chief policy officer at the International Federation of Reproduction Rights Organisations and a former official at the U.S. Patent and Trademark Office, has overseen the Office's AI-related initiatives. However, her tenure has been marked by a leadership dispute. On May 10, 2025, President Donald Trump claimed to have fired Perlmutter and replaced her with Paul Perkins as acting register. Perlmutter disputed the firing, noting that the Register of Copyrights is an employee of Congress, not the executive branch, and her claim to the role has been endorsed by courts. As of the report's release, Perlmutter remained in the position.
The Copyright Office is located in the James Madison Memorial Building of the Library of Congress, at 101 Independence Avenue SE, Washington, DC. The building, which opened in 1980, houses the Office's public reading room and administrative offices. The Office's staff includes examiners who review registration applications, attorneys who provide legal advice, and policy analysts who draft reports for Congress.
Recommendations for Legislation
The Copyright Office's key recommendations for deepfake regulation included:
- New federal law: Congress should enact a statute that creates a civil right of action for unauthorized digital replicas, covering both voice and likeness, with liability for direct infringement and for knowingly distributing such replicas.
- Duration of protection: The law should protect individuals for their lifetime plus a period after death, similar to copyright terms, but the Office suggested a shorter post-mortem period, such as 50 years, to balance public interest.
- Exemptions: The law should include exceptions for uses in news, commentary, criticism, and other First Amendment-protected speech, but these should be narrow and not allow for commercial exploitation.
- Safe harbor: Online platforms should have a notice-and-takedown mechanism, modeled on the DMCA, to limit liability for user-uploaded deepfakes, but with a requirement to implement reasonable measures to prevent repeat infringement.
- No preemption: The federal law should not preempt state right-of-publicity laws, but should provide a baseline that states can supplement.
The report also discussed the role of Artificial intelligence in creating deepfakes, noting that Generative AI models can produce realistic replicas with minimal human intervention. The Office acknowledged the rapid evolution of Machine learning techniques, including Deep learning and Neural network architectures, which have made deepfakes more accessible and harder to detect.
Reactions and Impact
The report was generally welcomed by industry groups and advocacy organizations, though some criticized it for not going far enough. The OpenAI and Anthropic companies, which develop large language models capable of generating text and images, had submitted comments to the Office during the inquiry, and they expressed support for a federal digital replica law. Some media companies and entertainment unions, such as SAG-AFTRA, had also lobbied for stronger protections, citing the threat of unauthorized deepfakes to performers' livelihoods.
The Copyright Office's recommendations have not yet been enacted into law. As of late 2025, Congress has held hearings on the issue, but no comprehensive federal digital replica statute has passed. Several states, including California, New York, and Tennessee, have enacted their own laws addressing deepfakes, but these vary in scope and enforcement. The Copyright Office's report is seen as a blueprint for federal action, and it continues to inform legislative efforts.
Relationship to Other AI Copyright Issues
The deepfake report is part of the Copyright Office's broader effort to understand the implications of AI for copyright law. In its earlier reports, the Office addressed whether AI-generated works can be copyrighted, concluding that human authorship is required, and whether training AI models on copyrighted works constitutes infringement, which it said depends on the specific use. The Office has also examined the use of Transformer (architecture) architectures, which underpin many modern AI systems, and the role of Large language models in generating text and images.
The Office's work on deepfakes intersects with its administration of the DMCA, which provides a notice-and-takedown system for online copyright infringement. The proposed digital replica law would extend a similar system to deepfakes, but with adjustments to account for the First Amendment and the unique nature of likeness rights. The Office has also considered the potential for Data Augmentation and other techniques to improve deepfake detection, but it has not made specific recommendations on technical standards.
Future Outlook
The Copyright Office continues to monitor developments in AI and deepfake technology. As of 2025, the Office has not issued additional reports on deepfakes, but it has indicated that it will update its recommendations as technology evolves. The leadership dispute over the Register of Copyrights could affect the Office's priorities, but its statutory functions remain unchanged. The Office's role in advising Congress on copyright policy is likely to remain central as AI-generated content becomes more prevalent.
In the absence of federal legislation, the Copyright Office's report serves as a reference for courts and policymakers. The Office has also updated its Compendium of U.S. Copyright Office Practices to reflect its policies on AI-generated works, and it continues to provide public information through its website and NewsNet service. The deepfake issue is part of a larger conversation about the intersection of Artificial intelligence and intellectual property, a topic that the Copyright Office is uniquely positioned to address.