Three Copyright Principles to Apply in AI Infringement Cases
As we quickly approach 60 copyright infringement lawsuits filed against AI companies, it is useful to understand three basic, incontrovertible copyright principles that will be applied by the courts in the context of these AI infringement cases.
Copyright Principle #1: Copyright Owners are Granted Various Exclusive Rights, Each of Which Must Be Considered Separately
Copyright owners are granted certain exclusive rights in their works under the Copyright Act. These rights include the right to reproduce (i.e., copy) the work, to prepare derivative works (typically referred to as the adaptation right), to distribute the work, to perform the work publicly, and to display the work publicly.
Each exclusive right is a standalone right that forms the cohesive bundle of rights—the copyright. As such, in an infringement analysis, courts must examine how the unique characteristics and differences between the exclusive rights are being implicated by the allegedly infringing act. This is especially true for the generative AI lawsuits, where the exclusive rights may be implicated multiple times and in multiple, different ways across different stages of the AI development and deployment process. Knowing which of the exclusive rights are being implicated and how they are being implicated, helps a court identify and specify the infringing acts. Without a baseline understanding the characteristics that make each of the exclusive rights unique, it is easy for a court to overgeneralize how a copyright is being infringed that can lead to incomplete, or even worse, incorrect, infringement analyses.
Copyright Principle #2: Identifying the Infringing Act(s) Is Key
When a third party engages in one or more acts that implicate one or more of the exclusive rights in a copyrighted work, that party is liable for copyright infringement unless either:
- there is an applicable exception in the copyright law that excuses the specific infringement(s) at issue, such as fair use; or
- that person is acting with the authorization of the copyright owner, such as when they have a license (or permissible sublicense) that permits them to engage in the otherwise-infringing act(s).
In the generative AI environment, there are at least two separate infringing acts that could result in liability for an AI company:
- infringement that occurs during the input process—often broadly referred to as ingestion or AI training; and
- infringement that occurs during the output stage—when a synthetic work is generated by the AI system.
During the input process, a copy of the copyrighted work is made by the AI company or developer. This act of copying infringes the copyright owner’s exclusive right to reproduce the work under Section 106(1) (absent a valid defense, such as fair use (more on that later)). Typically, a substantial similarity test is used to determine whether an infringing copy has been made by the third party. But, in these AI cases, exact replicas are being made so there is no real need to do a substantial similarity analysis since the original copyrighted work and the copy made by the AI company are by all accounts identical to one another (even if the resulting copy exists in a different medium or is subsequently altered).
At the output stage, the AI system might generate an output that is substantially similar to a copyrighted work. If that occurs, then there is a second, independent infringement of copyright owner’s reproduction right. The output might also infringe the adaptation right in Section 106(2) and the distribution right under section 106(3) (again, absent a valid defense) depending on the facts in the particular case.
It is important to distinguish between infringements that occur during the input stage and the output stage because the legal analysis necessary to determine whether an infringement has taken place is different. Some people mistakenly believe that in order to establish an infringement during the input stage, the copyright owner needs to establish substantial similarity between the ingested copyrighted work and AI-generated output and if no substantial similarity exists there is no infringement in this stage. That is incorrect. This argument ignores the fact that the right of reproduction enumerated in Section 106 of the Copyright Act is a stand-alone right—meaning that it can be violated through unauthorized copying regardless of whether there is a separate act of infringement at the output stage. Consequently, when determining whether an infringement occurs during the input stage, there is no need to compare the ingested work to the AI-generated output to determine if they are substantially similar because an identical copy of the work has been made during the input stage that violates the reproduction right in Section 106(1) of the Copyright Act. Thus, the substantial similarity test is not relevant to the legal analysis for ingestion.
As an aside, it is also important to recognize that for a copyright owner’s right of reproduction to be violated, there is no requirement that a copy of the work be downloaded, retained, or stored. The Copyright Act makes clear that a copy is made whenever a work is fixed and “can be perceived, reproduced, or otherwise communicated.” The Act further states that such perception can occur “either directly or with the aid of a machine or device.” The only requirement is that the copy exists for “more than transitory duration.” This is a very low standard that is easily met when AI companies ingest works for training. In fact, while some AI companies first tried to argue that ingestion doesn’t involve making a copy because there is no fixation, that claim has been completely debunked and AI companies have abandoned it as a defense.
Copyright Principle #3: There Are No Categorical Fair Use Exceptions
Because AI companies are indisputably making copies of works to train their AI systems, most of the almost 60 cases will ultimately come down to one question—Whether a defendant AI company’s training on copyrighted works at issue in a case constitutes fair use.
Determining whether a particular use qualifies for the fair use defense to infringement requires a fact-specific inquiry that is considered on a case-by-case basis. There are no uses that always, categorically qualify as fair use. While the preamble of Section 107 of the Copyright Act provides examples of uses that are more likely to qualify as fair uses, even those examples do not categorically qualify as fair use. That is no less true when copyrighted works are ingested for AI training. Put simply, blanket assertions that the ingestion of copyrighted works by an AI system should always qualify as fair use are legally inaccurate. A categorical exception for the broad notion of “AI training” would betray the flexible, fact-specific fair use analyses that our copyright system has long relied on.
Courts will need to evaluate fair use defenses involving AI systems the same way they evaluate fair use in all contexts: by applying the four factors set forth in section 107 of the Copyright Act to the specific uses at issue. When applying those factors to the typical commercial AI system’s ingestion of copyrighted works, courts should conclude that an AI system that generates competing works and serves the same purpose as the underlying works is unlikely to qualify as a fair use.
Under the first factor, a court must consider “the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes.” Part of the court’s analysis of the first factor is to consider whether the use is transformative, which is a use that adds something new, with a further purpose or different character that does not substitute for the original use of the work. In many cases, ingestion is unlikely to be a transformative use and the first fair use factor is unlikely to weigh in favor of fair use because (i) the use is commercial; and (ii) the output generated by these AI systems will often serve the same purpose as the works ingested. That is especially true in the case of AI systems related to music, images, and visual art.
Even if the use is held to be transformative, the Supreme Court recently made clear in Andy Warhol Foundation v. Goldsmith that whether a use is transformative is not dispositive of the question of fair use and is merely one of the considerations under the first fair use factor. In addition, under the fourth factor, when courts consider the extent of the “effect of the use upon the potential market for or value of” the works ingested by that system, they may conclude that such ingestion will have a significant adverse impact on the value and market for the copyrighted work. This is especially true when copyright owners have made licenses available in the market for AI training—something that is clearly happening today (see our blog on AI licensing and our list of AI licenses for more about the burgeoning AI licensing market).
When considering the first and fourth fair use factors, some people mistakenly think that the fair use defense requires that the AI-generated output be substantially similar to the ingested work. That notion is incorrect. As noted above, substantial similarity is part of the test for infringement. It plays no role in determining whether a particular use qualifies as a fair use. An AI-generated output can compete with, and therefore harm the market for, a copyrighted work that has been ingested even though the works at issue are not substantially similar to one another.
Conclusion
This discussion of fair use and AI merely scratches the surface. But this blog was intended to only discuss three basic copyright principles and how they would generally apply to the close to 60 AI copyright infringement cases. To go into a more detailed discussion of fair use would exceed that scope, so we will put a pin in this discussion for now and take up a more detailed discussion of fair use and AI at a later time. In the meantime, hopefully this blog post gave you a basic sense of how fundamental principles in copyright law should be applied to the close to 60 copyright infringement cases brought against AI companies.
If you aren’t already a member of the Copyright Alliance, you can join today by completing our Individual Creator Members membership form! Members gain access to monthly newsletters, educational webinars, and so much more — all for free!