Skip to Main Content
JMU Libraries logo .jmulib-logo-purple{fill:#450084;}
Loading

Copyright at JMU

A guide to copyright at James Madison University

Are the products of AI copyrightable?

A question that arises with increasing frequency is whether the products of Artificial Intelligence are, themselves, copyrightable. 17 U.S.C. § 102 speaks of "original works of authorship," a definition which limits copyrightable work to the products of human authorship. The United States Copyright Office examined the question in considerable detail and has affirmed this understanding of federal copyright law in a detailed report published in March 2013. The report further states that the crafting and development of prompts to guide or instruct generative AI in creating content does not give rise to any copyright in the resulting (AI-created) work. In short: if it's made by AI, at the current time it's not copyrightable - the work was born in the public domain.

Use of copyrighted works by generative AI

Generative AI, by definition, relies on existing works to create new content. At present, it isn't clear whether this use gives rise to a derivative work, and, in instances where the permission of the original copyright owner(s) has not been sought, if this then represents an unauthorized derivative work and hence a copyright violation. The question of whether the use of the existing work by AI is transformative fair use also remains unsettled. In all likelihood, we will not know that answers to these questions for several years, until current litigation makes it way through the federal courts and judgments are reached. In September 2023, a Congressional Research Service report noted: 

Plaintiffs have filed multiple lawsuits claiming the training process for AI programs infringed their copyrights in written and visual works. These include lawsuits by the Authors Guild and authors Paul Tremblay, Michael Chabon, Sarah Silverman, and others against OpenAI; separate lawsuits by Michael Chabon, Sarah Silverman, and others against Meta Platforms; proposed class action lawsuits against Alphabet Inc. and Stability AI and Midjourney; and a lawsuit by Getty Images against Stability AI.

An additional complication exists in terms of licensing and contract. Commercial publishers and vendors often seek to protect their products via end-user licenses, which may prohibit using those works to train AI models.

How to overcome these problems?

Until these questions are resolved by the courts, best practices include:

  • Training AI models on works that exist in the public domain
  • Training AI models on works that are covered by a Creative Commons license, and abiding by the terms of those licenses - this may include citations attributing the use of the original works in training the AI, attached to the end product the AI has made.