With more copyright applications being filed every day for works that incorporate AI-generated elements and over 9500 comments received in response to their Notice of Inquiry on the copyrightability of AI-generated works, the Copyright Office is being pushed to re-evaluate their initial policy on this subject, and the extent of protection for which AI-generated works are eligible. The questions get harder when an author is disabled and claims to be using generative AI as an assistive device, introducing a potential tension between Copyright Office policy and the Americans with Disabilities Act.
While the issues surrounding these questions remain unsettled, a recent case provides a window into where the Copyright Office could be headed.
The U.S. Copyright Office’s Position on the Registrability of AI-Generated Content
In March 2023 as the widespread adoption of generative AI first began to take hold, the Copyright Office published guidance pertaining to the registrability of “works containing material generated by artificial intelligence.” In that policy statement, the Copyright Office took a fairly bright line position that the output of generative AI is not the subject of human authorship, and so cannot be copyrighted. For our in-depth analysis of that policy and its potential implications, please see our article available here.
This policy generated significant industry reaction to what many found to be too simplistic an approach, and the Copyright Office issued a Notice of Inquiry to the public on this subject last fall. Over 9500 comments were received before the inquiry closed at the end of 2023, and many have been eagerly anticipating a revised policy statement since.
Elisa Shupe and AI Machinations: Tangled Webs and Typed Words
Our first glimpse into where that policy may be headed comes in the form of a Copyright Office decision issued in April, and which introduces new complexities related not only to AI, but also to disabilities. In October 2023, disabled U.S. Army veteran Elisa Shupe filed a copyright application for her self-published novel AI Machinations: Tangled Webs and Typed Words. Although the novel dons only Shupe’s pseudonym, Ellen Rae, she did not author the work without help. Shupe wrote the novel with the assistance of the publicly available AI chatbot ChatGPT.
Shupe’s initial application was rejected by the Copyright Office under its standing policy statement, which rejection Shupe appealed in January of this year. Shupe argued that, having been assessed as 100% disabled by the Department of Veteran affairs due to cognitive impairment related conditions, ChatGPT was merely an assistive tool—likening the use of ChatGPT to the use of a prosthetic limb by an amputee—and that the Copyright Office’s rejection based on her use of AI was thus discriminatory. Schupe also noted that ChatGPT did not do all of the work. Her appeal described the long hours she spent on her draft, her tailored prompting and the edits she made to the bot’s responses. According to Shupe, nearly every sentence was modified by her.
Ultimately, the Copyright Office granted Shupe a limited copyright for the compilation of AI-assisted sentences that comprised her book. The Copyright Office recognized Shupe as the author of the “selection, coordination, and arrangement of text generated by artificial intelligence,” but did not recognize her as the author of the text itself. In a way, this decision thus represents an extension of the Copyright Office’s earlier grant of a compilation registration for the selection and arrangement of AI-generated illustrations in the graphic novel Zarya of the Dawn.
What’s Next?
With the Shupe decision, the Copyright Office appears to be solidifying a tendency to treat works incorporating AI-generated elements as compilations for purposes of registrability, providing a possible window into where Copyright Office policy may be headed, at least as a next step. It will be interesting, in the wake of this decision, to see whether a wave of AI-based compilation applications are forthcoming, including for things like software for which generative AI tools are being broadly deployed. For those registrations which do issue, we expect questions will arise – and likely be litigated – over the scope of protection conferred by such compilation registrations as well as the associated presumptions of validity, including to the extent Copyright Office policy continues to shift. And this case begs the questions of if and how those with disabilities may use AI in the creation of copyrightable works, and how Copyright Office policy and the Americans with Disabilities Act may ultimately intersect.
While these questions remain uncertain, there is little doubt AI-generated content will continue to become more pervasive, and the efficiencies associated with the use of AI increasingly strong. Companies are well-advised to consider the evolving landscape of potential copyrights in AI-generated output when using models to create works they may seek to protect, as well as ways in which they may be able to leverage the present possibilities to secure available protections. And as those protections continue to shift, it becomes increasingly important to document what role AI did – and did not – play in the generation of works companies seek to protect and enforce.
