The emergence and rapid development of artificial intelligence raises interesting questions about copyright which generally fall into one of two categories. First, are works created by AI protected by copyright? Second, is the training of AI tools on works protected by copyright a violation of the rights of the original works or should it be considered fair use and thus not infringing?
According to the U.S. Copyright Office, for a work to be copyrightable, it must have human authorship. Works generated solely by AI without significant human input do not qualify for copyright protection. Merely prompting an AI tool does not qualify as significant human input. In instances where works are the combination of AI and human authorship, only the human authored elements are eligible for copyright protection, and those seeking to register their works with the U.S. Copyright Office are required to disclose and describe AI use in the work. For more information, including a report on the copyrightability of AI-authored works, see the U.S. Copyright Office on Copyright and AI.
The legality of using copyrighted works to train AI models is less clear and will be shaped by legal precedent and potentially legislative action in the coming years. Generally, using copyrighted material without permission could be infringing. However, there are exceptions, such as fair use, which is defined by the U.S. Copyright Office as non-infringing use without permission, particularly for the purpose of criticism, commentary, reporting, teaching, scholarship, or research. Fair use case law in the last several decades has often focused on whether a use is transformative. How courts will apply these ideas to training AI on copyright protected works is not yet known.