The recent explosion in popularity of generative artificial intelligence (AI), such as ChatGPT, has sparked a legal debate over whether the works created by this technology should be afforded copyright protections. Despite that several lawsuits on the subject have been filed, and the U.S. Copyright Office has recently issued guidance clarifying its position, that the bounds of copyright protections for works created using AI are not yet clearly defined and many questions remain unanswered. For now, it appears that copyright eligibility for such works depends on the extent of human involvement in the creative process and whether any use of copyrighted work to generate a new work falls within the purview of the fair use doctrine.

The analysis of this issue has been framed around two key aspects of the technology itself: input data and output data. Input data are the pre-existing data that human users introduce into the AI system that the system then uses to generate new works. Output data are the works ultimately created by the system—the finished product. Thus, copyright eligibility for AI-generated or AI-assisted works depends on whether the AI system’s use of copyrighted works as input data is permissible and whether the output data is itself copyrightable.

The question of whether AI systems can use copyrighted works to create new works has been analyzed under the fair use doctrine in the United States. In 2015, the Second Circuit determined that Google did not commit copyright infringement by scanning digital copies of copyright-protected books, creating a search function for the books that was accessible to the public, and displaying excerpts from the books for its Google Books service, because such conduct constituted fair use of the works. See Authors Guild v. Google, Inc., 804 F. 3d 202, 229 (2nd Cir. 2015). The Court reasoned that “[t]he purpose of the copying is highly transformative, the public display of text is limited, and the revelations do not provide a significant market substitute for the protected aspects of the originals.” Id.

While the Court’s decision in the Google Books case was limited to the specific facts at issue in that case, the decision supports the proposition that use of copyrighted works as input to a generative AI system in order to create a new work is protected under the fair use doctrine where the new work is highly transformative, minimally displays the original work, and does not act as a market substitute for the original work. Whereas, on the other hand, AI-generated works do appear to borrow heavily from copyrighted works or directly compete with them in the marketplace, they are far less likely to be considered fair uses of the original works and receive protection from copyright infringement claims. The exact bounds of the fair use doctrine in this context remain to be seen but will likely be addressed in the near future, as the universe of ways in which generative AI can use input data is wide-ranging.

The question of whether AI-generated or AI-assisted works are themselves copyrightable has recently been clarified by the U.S. Copyright Office. According to the Copyright Office, copyright eligibility for such works depends on the extent of human involvement in the creation of the work, which in turn is often dictated by how a given AI system operates. Because many of today’s most popular AI systems do not permit users to control how the system interprets prompts and generates material from input data, the Copyright Office likened use of such systems to “instructions to a commissioned artist” rather than direct control over the creative process. Thus, in 2022, the Copyright Office stated it will not register works created entirely by AI systems without any human input or intervention, because the copyright statute protects only works “created by a human being.” Second Request for Reconsideration for Refusal to Register a Recent Entrance to Paradise to Ryan Abbot (Feb. 14, 2022). And more recently, in March 2023, the Copyright Office notified a creator that it intends to cancel the copyright registration it previously granted for a comic book created with the assistance of AI because of its concern over the extent of human involvement in the creation of the work.

While the question of copyright protections for AI-generated and AI-assisted works remains somewhat murky, it is clear that works will not receive protection where they significantly display the original copyrighted work used as input data, compete directly in the marketplace with the original work, and are not transformative of original work, and additionally where they are created with minimal to no human involvement.