/>iOn December 16, 2024, the United States Copyright Office informed Congress that it was delaying release of Parts 2 and 3 of its report relating to copyright and AI until after January 1, 2025.
The Copyright Office’s report culminates a process beginning with a Notice of Inquiry in August 2023 relating to copyright law and policy issues raised by AI. The inquiry, however, was not limited strictly to copyright law, but sought input on whether the technology affected interests not currently protected by copyright law, specifically in areas involving rights closely aligned with publicity rights. Ultimately, the purpose of the Notice of Inquiry was, as described on the Copyright Office’s website, “to use this information to analyze the current state of the law, identify unresolved issues, and evaluate potential areas for congressional action.” The Copyright Office received over 10,000 stakeholder comments, including comments from artists, businesses, non-profits, trade associations, academics, and experts. The comments are a trove for people interested in learning how AI works and how it interfaces with copyright law.
Part 1 of the Copyright Office’s report, published on July 31, addressed what it refers to as digital replicas, e.g., video images and audio recordings that have been digitally created or manipulated to realistically but falsely depict an individual.
Part 1 details why existing state and federal laws—including laws relating to privacy, false light, and right of publicity—are inadequate for covering digital replicas, and it outlines proposed legislation that, as described in its December 16 letter to Congress, “recommends that Congress establish a federal right that protects all individuals during their lifetime from the knowing distribution of unauthorized digital replicas….” While digital replicas offer heightened risks of misinformation, fraud, and misappropriation, they also present new sources of value and expanded creative opportunities for profitable commercial exploitation. The Copyright Office described the need for legislation covering digital replicas as “urgent,” which may account for, at least in part, this issue being the focal point of Part 1 of the Copyright Office’s report.
Parts 2 and 3, which have not yet been released, will be of heightened interest to content creators and to individuals and businesses involved in developing and deploying AI technologies. Ultimate regulatory and legislative determinations could materially recalibrate the scope of ownership and protection afforded to works of authorship, and the stakes are extremely high.
“[W]hether the ‘work’ 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.) were actually conceived and executed not by man but by a machine.”
Part 2 of the report, which the Copyright Office expects to publish “after the New Year Holiday,” will address the copyrightability of AI-generated works, and more specifically, how the nature and degree of such use affects copyrightability and registrability. Current law is clear that to be copyrightable, a work must be created by a human. E.g., Thaler v. Perlmutter, 678 F.Supp. 140 (D.DC 2023), on appeal. However assistive tools are used in virtually all creation, from pencils to cameras to photo-editing software programs. In the context of registrability, the Copyright Office offered the following distinction in its March 2023 guidance: “[W]hether the ‘work’ 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.) were actually conceived and executed not by man but by a machine.” In Part 2, the Copyright Office will have an additional opportunity to explore these and related issues – this time with the advantage of the many comments offered through the Notice of Inquiry process.
Part 3 of the report, which the Copyright Office anticipates releasing “in the first quarter of 2025,” will focus on issues associated with training data. AI models, depending on their size and scope, may train on millions of documents—many of which are copyrighted or copyrightable— acquired from the Internet or through acquisition of various robust databases. Users of “trained” AI technologies will typically input written prompts to generate written content or images, depending on the model (Sora is now available to generate video). The output is essentially a prediction based on a correlation of values in the model (extracted from the training data) and values that are derived from the user prompts.
Numerous lawsuits, perhaps most notably the case that The New York Times filed against Microsoft and OpenAI, have alleged that the use of data to train AI models constitutes copyright infringement. In many cases there may be little question of copying in the course of uploading data to train the models. Among a variety of issues, a core common issue will be whether the use of the data for training purposes is fair use. Content creators, of course, point to the fact that they have built their livelihoods and/or businesses around their creations and that they should be compensated for what is a violation of their exclusive rights.
Developers and deployers of AI, on the other hand, argue that, among other things, the use of the content is “transformative” (a type of use that counts in favor of fair use); these stakeholders claim that the content is not being used for its expressive attributes (the core concern of copyright) but rather for information about the content that is ultimately translated into numerical and statistical values and patterns and that that information is being used to build a model and not for the purpose of the original creator. See, e.g., Thomson Reuters Enterprise Centre GMBH v. Ross Intelligence, Inc., 649 F.Supp.3d 467 (D. Delaware 2023) (case pending on motion for summary judgment, including issues relating to fair use, argued December 5, 2025).
Of course, these arguments beg the question. What really is the purpose of the developer’s use: to build the model or to create the output that may have similar use to the data used for training? Is the model really retaining only ideas and information about the expression and not the expression itself? Further complexity is added when considering (i) the nature of the output and the model’s capacity for and actual production of outputs that are similar to data used for training, and (ii) the definition of the markets relating to the inputs and outputs. Part 3 promises to give guidance on this and the myriad of other issues associated with training data.
Artificial intelligence has the capacity to add immeasurably to the panoply of images, text, video, and sounds that adorn and ultimately enrich culture – and in that sense fulfilling the constitutional goal of “promoting the progress of science and the useful arts.” While arguably the core concern of the Constitution is authorship, and not authors, we seem to care deeply about fulfilling human potential and about encouraging and protecting human creations. It will be interesting to see how these concerns emerge in parts 2 and 3 of the Copyright Office’s report, and ultimately in Congress.
