AI Developments at the U.S. Copyright Office in 2024

U.S. Copyright Office

“The art challenges the technology, and the technology inspires the art.” Such is the conundrum facing the U.S Copyright Office in this era of rapidly expanding generative artificial intelligence technology. Human creativity has been the cornerstone of copyright protection for original works of authorship ever since the U.S. Constitution recognized copyright as a fundamental right to be protected for limited times. But the tenet that originality exists only when a human is primarily responsible for creating works of authorship is currently in flux and subject to extensive debate. Nowhere is this tension more visible than within the Copyright Office itself, which has been grappling with the core issue of what defines human creation when sophisticated technology like generative AI plays a significant role in creating works of authorship under the direction of a human creator.

Monkeys may be highly intelligent, but in a 2018 decision the Ninth Circuit ruled that a photo selfie taken by an engaging macaque did not qualify for copyright protection because animals lack standing under the Copyright Act, which limits protection only for original works created by humans (Naruto [by PETA] v. Slater, 888 F.3d 418 (9th Cir. 2018)). The Copyright Office Compendium currently mandates that “[t]he U.S. Copyright Office will register an original work of authorship, provided that the work was created by a human being…. Because copyright law is limited to ‘original intellectual conceptions of the author,’ the Office will refuse to register a claim if it determines that a human being did not create the work.”

Works Created via Generative AI

Consistent with the seemingly engrained-in-stone principle that only human-created works are eligible for copyright protection, the Office has to date refused to register multiple works generated by AI platforms, regardless of the number of inputs (or “prompts”) applied by human creators to guide the resulting output of their creative concepts. In its 2023 “Copyright Office Registration Guidance: Works Containing Material Generated by Artificial Intelligence (Fed. Reg. Vol. 88, No. 51, 3/16/2023),” the Office made clear that if “a work’s traditional elements of authorship were produced by a machine, the work lacks human authorship and the Office will not register it.” As to registrability, the Guidance focuses on whether the work to be registered is “basically one of human authorship, with the computer [or other device] merely being an assisting instrument, or [whether] the traditional elements of authorship in the work (literary, artistic, or musical expression or elements of selection, arrangement, etc.)…are actually conceived and executed not by man but by a machine (Id.)?” Under the Guidance, applicants seeking registration “have a duty to disclose the inclusion of AI-generated content in a work submitted for registration and to provide a brief explanation of the human author’s contributions to the work. [S]uch disclosures are ‘information regarded by the Register of Copyrights as bearing upon the preparation or identification of the work or the existence, ownership, or duration of the copyright (Id.).’”

This Office policy led to a series of registration rejections starting in 2023 to date, where images were created using generative AI systems where humans provided the prompts that resulted in the creation of the final AI output. A work entitled Zarya of the Dawn received an Office letter on February 21, 2023, rejecting part of a registration application for a graphic comic to exclude all imagery generated by Midjourney AI technology, apart from the text created by the human author. In its letter, the Office stated: “[T]he images in the Work that were generated by the Midjourney technology are not the product of human authorship. Because the current registration for the Work does not disclaim its Midjourney-generated content, we intend to cancel the original certificate … and issue a new one covering only the expressive material that she created.”

In September 2023, the Copyright Office Review Board rejected (for the second time) an application to register a work called “Théâtre D’opéra Spatial,” which had been created by Jason Allen using the Midjourney generative AI platform. The Board concluded the work “contains more than a de minimis amount of content generated by artificial intelligence” to be eligible for copyright protection (Copyright Board proceeding No. 1-11743923581 (Sept. 5, 2023)). Allen had used at least 624 prompts (input instructions) to generate the initial image, then used Adobe Photoshop to correct flaws and finally Gigapixel AI to increase resolution and size.

In September 2024, Allen filed a declaratory judgment action under the Administrative Procedure Act against the Copyright Office in the District of Colorado, seeking to overturn the Office’s refusal to register “Théâtre D’opéra Spatial (Allen v. Perlmutter, No. 1:24-cv-02665 (D. CO.; complaint filed September 26, 2024)).”  In his complaint, Allen analogized his role as a creator using Midjourney, “who meticulously crafted the instructions,” to that of a film director “requesting multiple takes of each scene from a cameraman.”  Allen argues that “Midjourney, lacking autonomous creativity, simply executed the detailed guidance” he provided and that his “extensive effort and careful direction were crucial in creating the image, far surpassing the role of the AI tool.” In that respect, “Midjourney did not engage in any creative selection or arrangement of the image elements [but] simply followed the meticulously crafted instructions provided by the Plaintiff (Id. at ¶¶25, 28.).” Allen argues that it was his “selection, coordination, and arrangement of elements [that] resulted in the creation of the Work” even though it was “created ‘directly or with the aid of machine or device (Id. at ¶52.).’”

In December 2023, the Board rejected Ankit Sahni’s registration application for a work titled “SURYAST” in the style of Vincent van Gogh’s “The Starry Night,” which was created with AI painting software RAGHAV. The Board informed the applicant that his work could not be registered because he “exerted insufficient creative control” and “that the expressive elements of pictorial authorship were not provided by Mr. Sahni (Copyright Board proceeding No. 1-11016599571 (Dec. 11, 2023)).” Addressing the underlying AI technology, the Board emphasized that “it is the AI model, not its user, that ‘predict[s] stylizations for paintings and textures never previously observed,’ and that predictive function is tied to ‘the proximity of the [style image] to styles trained on by the model.’… Here, RAGHAV’s interpretation of Mr. Sahni’s photograph in the style of another painting is a function of how the model works and the images on which it was trained on—not specific contributions or instructions received from Mr. Sahni (Id.).”

Another court challenge to the Office’s refusal to register an AI-generated visual artwork received a cold reception by the D.C. District Court in August 2023 (Thaler v. Perlmutter, 687 F.Supp.3d 140 (D.D.C. Aug. 18, 2023)). The court reaffirmed established law that human authorship is required for copyright protection and granted summary judgement to the Copyright Office. The plaintiff had developed an AI program called the “Creativity Machine,” which he used to produce a work of art and, in his application to register the work, he identified the “author” as the Creativity Machine and himself as the copyright claimant. The court emphasized that human authorship remains essential and is “a bedrock requirement of copyright,” and that while the Constitution’s Copyright Clause was intended to incentivize the creation of original works of authorship, “[n]on-human actors need no incentivization with the promise of exclusive rights under United States law, and copyright was therefore not designed to reach them (687 F.Supp3d at 146; 147.).” As of the date of this article, the case is on appeal to the Court of Appeals for the D.C. Circuit.

Studying the Problem

As AI systems increasingly become more sophisticated, their outputs can require multiple creative decisions by human users submitting a series of text prompts, with each prompt submission being an incremental refinement to the image generation process. But unlike traditional digital imaging software, such as Adobe Photoshop, where users start with their own human-created visual works (independently protected by copyright) and then use the software to modify and enhance those images, generative AI systems directly produce original images and textual content based on prompts and their system training on large language models and massive image databases.

At the same time, generative AI providers are defending against numerous ongoing and expanding lawsuits alleging primary and secondary copyright infringement tied to AI system training on large language models that ingest massive amounts of online content, both text and imagery, and in some cases for the outputs themselves. This has pitted generative AI platforms and in some cases their investors, like Stability AI, OpenAI, Nvidia, Microsoft, and Anthropic, against a myriad of content providers and interest groups, like Getty Images, the New York Times, the Authors Guild and other human authorship individuals and class action groups. At the same time, AI is increasingly being used in entertainment markets to create or enhance content, which may be difficult to protect under copyright against wanton copying absence regulatory or legislative changes.

Against this backdrop, in August 2023, the Copyright Office issued a Notice of Inquiry for an Artificial Intelligence Study, to allow the Office to “analyze the current state of the law, identify unresolved issues, and evaluate potential areas for congressional action.” The Notice required comments to be submitted by October 30, 2023, and reply comments by December 6, 2023. Over 10,000 comments were received by the Office, highlighting the significance of generative AI in society and the myriad interest groups it impacts. To assess whether legislative or regulatory action is warranted, the Notice invited comments on issues respecting “use of copyrighted works to train AI models, the appropriate levels of transparency and disclosure with respect to the use of copyrighted works, and the legal status of AI-generated outputs.” The scope of the Inquiry included both visual (text, visual image, video) and audio outputs “emulating a human voice … that would be considered copyrightable if created by a human author (Id.).”

On July 31, 2024, the Office published Part 1 of its AI Report, addressing the subject of digital replicas, which also has been the focus of ever-increasing state legislation and potential federal legislation (See, e.g., “Nurture Originals, Foster Art, and Keep Entertainment Safe Act of 2024” (a/k/a NO FAKES Act) H.R.9551 and S.4875, both introduced in 2024 with bipartisan support)). The Report defines a digital replica as “a video, image, or audio recording that has been digitally created or manipulated to realistically but falsely depict an individual.” Part I assesses whether “existing laws provide sufficient protection against unauthorized digital replicas or if new protection is needed at the federal level (Id. at p. iii).” Due to a plethora of unauthorized digital fakes and increasingly fragmented state-level legislative initiatives to curb related abuses, the Office concluded that new federal legislation was urgently needed in order to minimize ongoing harm to private individuals, as well as harm in the fields of entertainment and politics. Protections should extend to ordinary individuals, not just celebrities whose persona has commercial value, and last for individuals’ lifetimes, with any post-mortem protection being limited in duration. The Report recommends imposing liability only for the distribution or making available of unauthorized digital replicas, but not the act of creation alone, which arguably would omit potential liability for bad actors who only “create” unauthorized digital replicas but know they will be distributed with harmful intent. Part I also supports imposing secondary liability while providing a safe harbor defense to online service providers who remove unauthorized digital replicas after receiving statutory notice or otherwise learning that such replicas are unauthorized.

Other legislative initiatives supported by the Office include implementing a licensing program for individuals, who would be able to monetize their digital replica rights but not assign them outright; providing remedies for both injunctive and monetary relief; and balancing free speech concerns rather than dictating categorical exemptions. Notwithstanding the potential for conflicting state laws governing digital replicas, the Office does not advocate full federal preemption, insofar as rights of publicity are creatures of state statutory and/or common law. Instead, the Office encourages any new federal statute to provide a “floor” where states would then be free to add additional protections. In this context, the Office made clear that Section 114(b) of the Copyright Act, which defines limitations to the exclusive rights of owners of copyrights in sound recordings, does “not preempt or conflict with laws restricting unauthorized voice digital replicas.”

Much-Needed Recommendations to Come in the New Year

Notwithstanding the very serious concerns raised by unauthorized digital fakes in our society and the critical need for legislation, it is the Office’s Part II Report that is even more anticipated because it will focus on “the copyrightability of materials created in whole or in part by generative AI, the legal implications of training AI models on copyrighted works, licensing considerations, and the allocation of any potential liability.” As of the date of this article, Part II had not yet been released by the Office, but a follow up article will be forthcoming by the author once it is made public.