Copyrighted Works Used in Generative AI Training – Fair Use or Infringement?

With the rise of the use of generative AI training comes the risk of potential lawsuits from copyright owners whose works have been used without their authorization. The US Copyright Office is stepping up to provide guidance to keep up with this evolving technology and help define what constitutes fair use.

On May 9, 2025 (the day before the Register of Copyrights was fired by the White House)  the United States Copyright Office released Copyright and Artificial Intelligence, Part 3: Generative AI Training Pre-Publication Version with respect to uses of copyrighted works in Generative AI Training. This is the third report on AI and copyright released by the United States Copyright Office.

The report outlines four factors in discussing the determination of fair use, as follows:

Factor one: The transformativeness and commerciality of the infringer’s use. The transformativeness looks at how the AI model is deployed and its purpose – whether it is merely a substitute for the original, serving the same purpose, or if it differs from the original and adds something new. The commerciality looks at potential unfairness of the use of copyrighted works for financial benefit (not merely monetary gain) of the infringer. The report also looks at whether the infringer had lawful access to the work.

Factor 2: The quantity of material used and its quality and importance. The report considers more creative or expressive works are less likely to be considered fair use, while use of more factual or functional works is more likely to be considered fair use. A work that is factual but also contains expressive material may well not be considered fair use.

Factor 3: The amount and substantiality of the portion of the copyrighted work used. Use of an entire work is unlikely to be considered fair use, even for training, and even using a small portion, if it is considered to be the meat of the work, may also not be considered fair use.

Factor 4: The potential for market harm. This factor has been found to be the most decisive in fair use on several occasions by the Supreme Court. This includes not only harm to the original work but also to the market for derivative works. Examples of market harm include lost sales of the original work, market dilution (imitations of the style or feel of the original works possibly diluting the value of the original), and loss of licensing for similar works as the wealth of computer-generated content makes it harder for human authors to compete.

TAKEAWAY: The report is so far the only guidance from the Copyright Office on the topic of AI training models with copyrighted materials. Based on the guidance, the issue as to what constitutes fair use will no doubt continue to be challenged and further defined as technology and the relevant law evolve and consideration of the four factors above will assist in the determination of what is fair use and aid in the avoidance of lawsuits based on generative AI training infringement.

A threshold issue, however, is whether the report will even become and remain official policy given the abrupt termination of the Register.    

Add a comment

Type the following characters: six, foxtrot, three, november, whisky, hotel

* Indicates a required field.

Subscribe

Recent Posts

Contributors

Archives

Jump to Page

Necessary Cookies

Necessary cookies enable core functionality such as security, network management, and accessibility. You may disable these by changing your browser settings, but this may affect how the website functions.

Analytical Cookies

Analytical cookies help us improve our website by collecting and reporting information on its usage. We access and process information from these cookies at an aggregate level.