Earlier this year, the U.S. Copyright Office released part two of its artificial intelligence (AI) report addressing the copyrightability of outputs created using generative AI. This new report is largely consistent with the Copyright Office's earlier registration decisions and guidance. The central question the Copyright Office poses regarding the protectability of AI outputs remains whether the AI tool was used merely as an assisting instrument or whether the traditional elements of authorship were conceived and executed by the tool rather than a human. The Copyright Office provides more explanation as to its reasoning and provides some additional guidance on when works may be protectable despite having been created with the aid of AI. However, it continues to set a high bar to meet the human-authorship requirement, and the exact boundary between protectable and unprotectable works remains undefined.
Note that although the Copyright Office's position is not legally binding and a court could rule differently, its views can be considered persuasive authority, and the Copyright Office determines what works achieve copyright registration. Moreover, until there is existing caselaw or legislation is passed (which is unlikely), the Copyright Office is providing the only available guidance on this issue. As such, the Copyright Office's position may carry additional importance as a potential indicator of how the issue of copyrightability of AI-generated works should be resolved.
Human Authorship Requirement Is Reaffirmed
The report starts by reiterating the Copyright Office's long-standing position that copyright protection requires human authorship. It cited, among other things, the Copyright Clause of the U.S. Constitution (which secures rights to "authors") and a recent district court case that confirmed that copyright does not extend to materials that are solely AI-generated without any human involvement. The report then turns to the question of the nature of the contribution necessary to qualify as authorship. The Copyright Office concludes that the resulting work is copyrightable only if it contains sufficient human-authored expressive elements. It ultimately focuses on the concept of control to determine this and emphasizes that merely providing detailed suggestions and directions are insufficient, as they are unprotectable ideas.
Assistive Uses of AI Systems To Enhance Human Expression Do Not Affect Copyright Protection
The Copyright Office makes an important distinction between a person using AI as a tool to assist in their creation of a work and AI being used as a replacement for human creativity. It notes that the distinction between these two modalities depends more on how a system is being used rather than its inherent characteristics.
The Copyright Office concludes that "assistive uses that enhance human expression" do not limit copyright protection. The report provides examples of tools that age or de-age or that remove unwanted objects or crowds from a scene as ones that would not affect the availability of copyright protection. The report does not suggest that these AI-generated changes themselves would be excluded from the copyright, and it therefore appears there is likely no need to disclaim them in an application.
The report distinguishes this from situations where an AI system makes "expressive choices," which will require further analysis. The Copyright Office does not, however, explain how a tool that automatically fills in the space that an unwanted object once occupied or determines what someone may look like as they age provides its human users with any greater degree of control than other AI tools that fulfill the directions of its user or why the choices made by such a tool are any less "expressive."
Copyrightability Depends on the Nature of Human Contribution
Where an AI system does make "expressive choices" and is not just an assistive tool, the report says that further analysis about human authorship is required. The report focuses on three types of human contribution to AI-generated outputs: (1) human-provided prompts that instruct an AI system to generate an output; (2) human-created expressive inputs that can be perceived in AI-generated outputs; and (3) modifications or arrangements by a human of AI-generated output.
Instructional Prompts
While the Copyright Office provides examples of how the last two types of human contributions can support an authorship claim, it concludes that instructional prompts alone provide insufficient human "control" over outputs for users to be regarded as the author of such output (at least as current AI tools operate). To support its focus on "control" as the determining factor of copyrightability, the Copyright Office cites caselaw involving joint authorship, which looks to the relative control exercised by each party and the amount of control necessary to claim authorship.
The Copyright Office equates prompts with giving instructions to a commissioned artist and concludes that things like providing detailed suggestions and directions, without influence over how those directions are executed, are insufficient for authorship as they are merely unprotectable ideas. However, the report does not fully explain why highly detailed prompts, at least along with extensive refinement of the output, would not provide sufficient influence over the final product so that the output would be deemed copyrightable. In one of the cases the Copyright Office cites, Community for Creative Non-Violence (CCNV) v. Reid, the court of appeals remanded the case (which involved a commissioned sculpture) to the district court to determine if the hiring party, CCNV, could also be an author, but the case settled before a decision was reached. While the court of appeals recognized that copyright does not extend to ideas, it noted that the sculpture was more than an abstract idea. It found that the sculptor's claim of sole authorship was prominently undercut by the district court's findings that CCNV was the motivating factor in creating the sculpture, conceived the idea in starkly specific detail, and directed enough of the sculptor's effort to assure that, in the end, he had produced what CCNV, not the sculptor, wanted.1
Someone who uses detailed prompts in an AI tool to conceive the idea in similarly specific detail and directs the AI tool (e.g., through an iterative refinement process) to achieve the desired result could similarly argue that he should also be considered an author. However, the report tries to distinguish human collaboration from directing an AI tool, saying that in a human-to-human collaboration, the hiring party is able to oversee, direct, and understand the contributions of a commissioned human artist (which, depending on the facts, could result in joint authorship). The Copyright Office does not clearly explain why an AI user cannot be seen as overseeing and directing the process if they can show they exert sufficient influence over the final result.
The Copyright Office also concludes that gaps between what was directed by prompts and the resulting outputs evidence a lack of control. The report provides the following example:
The Copyright Office points out that, while the output reflects portions of the prompt (e.g., a bespectacled cat smoking a pipe) but not others (e.g., a highly detailed wood environment), where no instructions were provided, the AI system filled in the gaps (e.g., the breed of the cat). Similarly, they claim the fact that identical prompts can generate a variety of different outputs further indicates a lack of human control. Although the Copyright Office did not seem concerned about an AI system filling in gaps in an "assistive use" or that an assistive use can also result in a variety of outputs, with respect to prompts, they conclude these factors demonstrate a lack of control.
The Copyright Office, addressing criticism of its focus on control and predictability, acknowledges that having some uncontrolled or random elements in an output does not always negate authorship. For example, the report discusses comments that drew analogies to a Jackson Pollock painting or wildlife photograph, which are recognized as copyrightable even if the author doesn't control where the paint falls on the canvas or when the wildlife will move into the frame of the photo. The Copyright Office explains that in such examples, the human author is "principally responsible" for the execution of the idea and the determination of the expressive elements that appear in the end result. Human artists and photographers control the staging, colors, depth, and more of the work they create, even when, for example, the subject of the piece may be variable. They distinguish this from an AI-generated work, claiming that the human is not principally responsible for the resulting output from a prompt, and conclude that some element of randomness does not eliminate authorship. The report notes that the author must be able to constrain or channel the processing of the source material and that the issue is the degree of human control rather than the predictability of the outcome (a helpful clarification from their prior guidance).
The Copyright Office does leave the door open to the idea that a prompt-based system, in theory, could someday afford a human the requisite level of control over an output to warrant copyright protection, although it provides little guidance as to how such a tool would operate. The Copyright Office does suggest that the AI system's contribution would have to be merely rote or mechanical for the user to be the author of the output, citing cases involving joint authorship. It is not entirely clear, however, how the Copyright Office distinguishes cases like CCNV, which recognized that a commissioning party could potentially be a joint author, even where the hired party's contribution was not rote. It is possible, however, that the Copyright Office views AI differently since joint authorship would not be applicable.
Expressive Input
For expressive inputs, the Copyright Office states that where a human inputs their own copyrightable work as a form of prompt and the work is perceptible in the output, the perceptible human expression in the output would remain copyrightable. The report provides an example (below) of a work that was submitted to the Copyright Office and notes that the elements of the input illustration that were perceptible in the output (e.g., the outline of the mask; the position of the nose, mouth, and cheekbones; the arrangement of the stems and rosebuds; and the shape and placement of the four leaves) were protectible. The applicant disclaimed any "non-human expression" appearing in the final work (e.g., the realistic, three-dimensional representation of the nose, lips, and rosebuds and the lighting and shadows), and the Copyright Office limited the registration to the "unaltered human pictorial authorship that is clearly perceptible in the deposit and separable from the non-human expression that is excluded from the claim."
This decision was not particularly remarkable, as the illustration is clearly a copyrightable work, and by limiting protection to the elements perceptible in the output, the Copyright Office is still not recognizing any computer-generated elements as protectable (even though it acknowledged that this type of input contributes more than just intellectual conception and limits the range of outputs).
It is notable, however, in that it is a markedly different result from the Copyright Office Review Board's decision upholding a refusal to register an image titled "Suryast" that involved the use of a photograph as an expressive input. Suryast was created using an AI style-transfer tool that applied a style image (Van Gogh's Starry Night) to the author's base photograph, applying a human-selected transfer strength value (which determined the amount of style transfer to be applied). The Copyright Office distinguished these applications, saying that unlike Rose Enigma, the output did not clearly show the copyrightable work input by the applicant. However, elements of the photograph do appear visible in the Suryast image. It is also not entirely clear why the Copyright Office does not see this as an assistive use that enhances human expression (which, as noted above, it says does not detract from copyright protection).
Arrangement/Modification of an Output
The third type of human contribution to AI-generated outputs that the report discusses is modifying or arranging AI-generated content. From its earliest decisions on registrations involving AI-generated works, the Copyright Office has consistently recognized that even if an AI-generated output isn't protectable, human modifications to that output could, in some cases, be sufficiently original to merit copyright protection. It has also recognized that the selection and arrangement of AI-generated materials in a sufficiently original way (or the combination of AI-generated work with copyrightable work) could result in copyright protection for the work as a whole (but excluding any AI-generated elements). It is worth noting that the standard for originality was set in the Feist case and is quite low—just a modicum of creativity.
The report acknowledges these principles and states that it will review such instances on a case-by-case basis, and where the modifications or arrangements rise to the minimum standard of originality under Feist, the output—not the AI-generated content alone—will be copyrightable.
Copyrightability Questions Can Be Resolved Under Existing Law
The Copyright Office concludes (in line with its prior guidance) that existing legal doctrines are sufficient to resolve questions of copyrightability for AI-generated works such that new legislation is unnecessary. The Copyright Office states that the case has not been made for additional protection for AI-generated works—such as sui generis rights—beyond what is already provided by existing law. In support of this conclusion, it notes that because the Copyright Clause requires human authorship, copyright law cannot serve as the basis of protection for works that do not satisfy the human authorship requirement. In response to opinions that incentives to create new works are needed, the Copyright Office disagrees. It argues there are significant existing incentives for developers of AI programs and systems, including patent, copyright, and trade-secret protection for the machinery and software, as well as potential funding. With respect to users of AI systems, it notes that the creation of works using AI requires substantially less time and effort than most human-created works and that humans receive copyright protection to balance against the cost of creating those works. Additionally, the Copyright Office notes that it is not convinced that additional incentives would promote progress and expresses concern about the effect of AI-generated material on human authors.
Takeaways
- The Copyright Office continues to take a fairly narrow view of copyright protection for AI-generated content and to focus on control over output as a key factor.
- Assistive uses of AI that merely enhance human expression (such as de-aging or removing background objects) were found not to affect the availability of copyright protection.
- The report recognizes protection where a copyrightable work is used as input to the extent that the work remains perceptible in the output.
- Many questions still remain, such as what the Copyright Office would qualify as an "assistive use" and what is sufficient control to meet the human-authorship requirement.
- Documentation of human involvement with the AI content generation process will be increasingly important for creators seeking copyright protection. This includes documentation as to how AI tools are used and how they influence and control the generated output, as well as documentation as to how outputs were modified and how they were selected, arranged, and combined with other elements to create the final product.
Footnote
1. Cmty. for Creative Non-Violence v. Reid, 846 F.2d 1485 (D.C. Cir. 1988).
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
