Artificial intelligence (AI) has dominated the headlines over the last several years. As technology has continued to advance, computers and robots have progressed from merely assisting human beings with common tasks to making increasingly more sophisticated decisions on their own. Generative AI bots now engage in many creative endeavors, such as writing poems, creating digital artwork, and editing images. While courts have consistently held that copyrights only protect works created by human authors, legislators, government officials, and other legal minds are grappling with the scope of copyright laws as the line between human and AI creation continues to blur. In an effort to provide guidance to its applicants, the Copyright Office has recently released a policy statement1 regarding the registrability of AI-assisted works. While this guidance brings much needed clarity to the copyright registration process, questions remain as to the legality of AI-generated works and the process to create them.
Background
In order for a work to be protected by a copyright, it must meet the minimum statutory standards of being an “original work[] of authorship fixed in any tangible medium of expression.”2 It is long standing Supreme Court precedent that the “authorship” requirement is limited to human creativity.3 The Copyright Office as well as federal courts have consistently refused copyright protection for works created by animals, forces of nature, divine beings, and mechanical processes.4 The Compendium of the US Copyright Office Practices specifically addresses the role of technology, stating that, “the [Copyright] Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human creator.”5
While non-human authorship is beyond the scope of copyright protection, technology has regularly played a role in the creative process. Artists have wielded tools, such as cameras, electronic musical instruments, and image-editing software, to produce copyright-protected works. However, artificial intelligence is posing unique challenges to this legal framework due to the technology’s level of participation in the creative process. Software programs are no longer merely acting as tools to assist human authors but are now often engaging in creative decision-making and content generation. Software programs are fed massive datasets of images, videos, and text, compiled by crawling the internet and other databases, in order to train the AI system to identify patterns. These programs can then draw upon their training and output original expressive content. For example, software bots have viewed entire catalogues of television shows and movies to write new scripts, resulting in social media sensations and entirely bot-generated web series.6 As technology has become increasingly entwined in the creation of art, confusion has arisen as to the scope of copyright protection for such works.
The Copyright Office Issues Policy Statement Regarding Registrability of AI-Assisted Works
On March 16, 2023, the Copyright Office issued further guidance on how it evaluates copyright applications for works that involve artificial intelligence. The policy statement seeks to clarify the Office’s policies and practices and does not issue any new rulemaking.
In this statement, the Copyright Office lays out its examination process for evaluating applications for works that include AI assistance in the creative process. As an initial matter, the Office maintains its position that works that are entirely the work of non-human authors, such as artificial intelligence, cannot be protected by copyright. The Office specifically rejected the notion that prompts, queries or commands to create a specific type of image or text made by a human to AI technology are sufficient to meet the authorship requirement. Viewing the prompts as analogous to the instructions provided to a commissioned artist, the office concluded that works cannot be copyrighted if the machine is making the decisions regarding expressive content. On the other extreme, the Office likewise reiterated its position that copyright protection extends to works that are wholly the product of human creative decision-making with mere mechanical assistance from technology, such as cameras and editing software.
However, special consideration is given to applications that combine human and AI work product. Specifically, works that are the result of “sufficient human authorship” may be registrable with a limited scope of protection.7 When examining an application that implicates artificial intelligence, the Copyright Office will separate those elements that were authored by a human from those that were generated by a computer. Registration will only extend to that portion of the work that was the result of human creativity. Protection can extend to works that were created by a computer but involve subsequent creative input from a human author, such as compilations or derivative works of AI-generated materials, but only to the man-made creative additions and not the underlying AI-derived works. The Office cautioned that this is a fact specific inquiry and will be evaluated by its examiners on a case-by-case basis.
To illustrate its process, the Copyright Office discussed its recent approval for registration of the graphic novel Zarya of the Dawn, which contains both human-authored and AI-generated elements.8 The comic book was written and compiled by a human author, Kristine Kashtanova, but contains illustrations created by an AI web service. Although the original registration was cancelled due to failure to disclaim AI-generated materials, the Copyright Office subsequently approved the graphic novel for registration but limited the scope of the copyright to the human-generated elements, namely, the text, storyline, and arrangement and selection of AI-produced images.
The limits of what constitutes sufficient human creativity will be an ongoing question as artists continue to push the boundaries of human and AI collaboration. Illustrating the difficult application of this standard, Kristine Kashtanova has recently filed a new application for a digital art piece entitled “Rose Enigma,” which was created by an AI-generative model based on a sketch and detailed prompts regarding subject matter, composition, and artistic style by the human author.9 The Copyright Office will be required to expand upon its examination procedures and begin drawing lines as to what is and is not sufficient human involvement as this application and many more come up for review.
Legal Questions Linger Related to AI-Generated Content
While the Copyright Office’s new guidance provides some clarity for applicants seeking to register their copyrights, substantial legal questions remain. First, this new guidance only relates to the copyright registration process. While registration has significant benefits for owners and is a prerequisite for filing suit for copyright infringement in the US,10 registration is not necessary for copyright protection to adhere. Copyright protection begins upon creation of the work.11 Thus, the Office’s policy statement does not address unregistered works and how copyright owners can protect their rights in the midst of a technology boom. As more AI-generated and -assisted content, both registered and unregistered, enter the market, it will be increasingly difficult for copyright owners to detect and police unauthorized uses of their works by software programs.
Second, many copyright owners have raised concerns regarding potential infringement by AI software in the content generation process. Several lawsuits have been lodged by artists and copyright owners for the unauthorized use of their works by software developers and AI programs.12 These lawsuits claim that the AI developers are scraping online platforms for images, videos, text, software code, and other copyright-protected works in order to train their software. The AI then relies upon these copyrighted works as (at best) inspiration or (at worst) as a source of material to develop new creative content. The copyright owners claim that such conduct amounts to copyright infringement. Software and AI developers defend against these infringement claims by citing the fair use doctrine to justify their use of these copyright-protected works. “Fair use” is a judge-created doctrine applying a four factor test; its conclusion in any given case is difficult to predict. Appellate courts – including the US Supreme Court – have rejected efforts to create bright-line rules or presumptions that parties could use to guide their behavior.13
Interestingly, these lawsuits implicate issues beyond the scope of copyright law. For example, Getty Images in its lawsuit filed against Stability AI, alleges that the AI-generated images infringe upon the company’s trademarks.14 Getty Images offers access to millions of photographs and images and protects its digital assets by including a watermark of the company’s trademark on each image. As stock image services, like Getty Images, are one of the largest sources of material used in AI training, software programs sometimes incorporate reproductions of these watermarks in computer-generated images. Getty Images argues that the inclusion of its trademarks in connection with poor quality AI-generated content tarnishes and dilutes its brand on the market. Other third parties may well continue this trend of challenging the legality of AI programs for a variety of claims related to misuse and misappropriation of intellectual property, and breach of website terms and conditions.
Because each of these cases are in the early stages and no decisions have been handed down, it is difficult to predict how the law will develop in this area. As in the famous Grokster copyrighted music streaming case,15 courts may determine that these AI programs are infringing, which would require sweeping and systemic changes in the software development industry, including the possible development of text and image licensing schemes similar to that which now applies to music streaming. On the other hand, courts may side with the developers and rule that such conduct is protected by the fair use doctrine, as AI-generated content is transformative of the underlying works. It is also possible that courts will find a middle ground in which distinctions are drawn based on the nature of the use, such as whether the copyrighted works are used merely to train the software or are used to generate new content. Courts may also consider whether the AI was developed to mimic and thus compete directly against the works of specific artists and authors.
Rather than wait for a judicial or legislative solution, some platforms have taken the initiative to address these concerns on their own. Some online artist communities have banned the upload of AI-generated content to their servers. In the fall of 2022, Getty Images was the first major platform to launch a policy prohibiting the upload of software-created works due to concerns over copyright infringement.16 Many websites have begun including prohibitions on scraping content and datamining in their terms and conditions and seeking enforcement through breach of contract actions.17 As the law lags behind the speed of technology, individuals and companies will continue to search for creative solutions to limit the unauthorized uses of their works by artificial intelligence.
Considerations for Content Creators
While the future of AI in the copyright space may be unsettled, individuals and businesses can take some lessons from the current state of the law to evaluate their content creation processes and intellectual property protection regimes.
Document the role and output of technology during the creative process.
Individuals and companies utilizing artificial intelligence in the creation of artistic works and other expressive content should carefully keep records of the involvement of both human and computer actors in the creative process. When applying to register a copyright, the Copyright Office requires that applicants identify when a work contains AI-generated material and to distinguish between the contributions of the human author and the computer. If an application identifies that AI-generated content is part of the work, copyright examiners have the ability to contact the applicant to request additional information. Having accurate documentation can be critical to establishing sufficient authorship and ultimately may be the deciding factor as to whether a registration issues.
Carefully draft copyright applications.
When preparing a copyright application, applicants should be careful to claim only those elements of the work produced or contributed by a human author. The standard copyright application includes data fields to disclaim computer-created materials from the scope of the copyright application. Any applications already filed or registered should be corrected if AI-generated materials were inadvertently claimed. Accurately claiming only those human-authored elements of a work is necessary to avoid challenges to the validity of a copyright registration and potential cancellation.
Seek permission or licenses whenever possible before using copyrighted materials.
Since it is unclear whether the use of copyrighted materials to train AI systems is actionable as infringement, those engaged in developing, training, and operating artificial intelligence systems should carefully consider how and to what extent they are using third-party content. Due to the highly fact specific nature of the doctrine, reliance on fair use can be a risky proposition. Individuals and companies should consider seeking authorization from copyright owners before using protected materials.
Weigh in on the development of new rules and regulations.
Throughout the spring, the Copyright Office will be hosting listening sessions for artists and other stakeholders to discuss the role of AI and the scope of copyright law. Later in the year, the Copyright Office plans to solicit public comment on specific topics related to artificial intelligence in order to facilitate further policy statements and potential rulemaking. Information regarding these events and other updates from the Copyright Office related to AI can be found on their newly launched webpage (https://copyright.gov/ai/).
Conclusion
Artificial intelligence will continue to play a greater role in all facets of society and our daily lives. As technology becomes more integrated in the creative process, the law will be playing catch up to conform to our new reality and craft rules and regulations to protect the interests of intellectual property holders.
___________________
1 Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16,190 (Mar. 16, 2023).
3 See, e.g., Burrow-Giles Lithographic Co. v. Sarony, 111. U.S. 53 (1884).
4 See, e.g., Urantia Found. v. Kristen Maaherra, 114 F.3d 955 (9th Cir. 1997); Kelley v. Chi. Park Dist., 635 F.3d 290 (7th Cir. 2011); Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018); Stebbins v. Polano, No. 21-cv-04184-JSW, 2022 U.S. Dist. LEXIS 121919 (N.D. Cal. July 11, 2022).
5 U.S. Copyright Office, Compendium of the U.S. Copyright Office Practices § 313.2 (3d ed. 2021).
7 See supra note 1, at 16,192.
12 See, e.g., Doe 1 v. Github, Inc., No. 4:22-cv-6823 (N.D. Cal. filed Nov. 3, 2022); Andersen v. Stability AI Ltd., No. 3:23-cv-00201 (N.D. Cal. filed Jan. 13, 2023).
13 Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994)(rejecting the circuit court’s conclusion that "every commercial use of copyrighted material is presumptively ... unfair .... ").
14 Getty Images (US), Inc. v. Stability AI, Inc., No. 1:23-cv-00135-UNA (D. Del. filed Feb. 3, 2023).
15 MGM Studios Inc. v. Grokster, Ltd., 545 U.S. 913 (2005).
17 See, e.g., hiQ Labs, Inc. v. LinkedIn Corp., No. 17-cv-03301-EMC, 2022 U.S. Dist. LEXIS 237258 (N.D. Cal. Oct. 27, 2022).
[View source.]
