Subscriber Benefit
As a subscriber you can listen to articles at work, in the car, or while you work out. Subscribe Now
How many times have you said, heard, or read the abbreviation “AI” this year? Thousands? Tens of thousands?
I am sure you have experienced the endless stream of public announcements about companies investing in AI or integrating AI into many services you and your clients experience every day.
Most of the AI you are hearing about or using is the “generative” variety, a system that generates content in response to a prompt. The new output can resemble patterns, style and content from a training dataset.
As a day-to-day user of an AI system, you probably do not know the scope of what goes into training these systems.
The problem many of these AI companies have encountered is that their enormous training datasets contain enormous amounts of copyrighted content. Many copyright owners noticed and filed dozens of lawsuits across the country, some of which are now proceeding as class actions.
It has taken a few years, but we are starting to get substantive decisions on whether, and which aspects of, a generative AI system is “fair use” exempt from copyright infringement liability.
As is common in fair use, the factual details of each case matter greatly to the application of the fair use doctrine and the multi-factor test set out in 17 U.S.C. § 107.
Early in the year, an AI service was found to be an infringement because the output was directly competing with the copyright work. Then, two decisions over the summer agreed that training generative systems and their use was transformative and qualified for fair use.
In one of those cases, however, additional pirated copies of the copyrighted content were found not subject to the fair use defense, leading to a historic settlement with authors. So far, none of these cases have reached an appellate decision or are positioned for the Supreme Court to intervene.
In addition to claims against the training aspect of AI systems, rights owners also assert that the outputs are infringements, either by direct action of the AI company or indirectly by enabling and contributing to infringement by users. Many of these cases are still in earlier stages. Despite a handful of settlements this year, the number of lawsuits is still growing.
Against the backdrop of all these disputes, recent announcements focus on deals. Disney is licensing hundreds of characters to OpenAI for use in consumer-prompted outputs as well as taking an equity stake in the company. Several large tech companies and AI platforms now have agreements with many publishers for licensed model training, direct use/display of source content in outputs, and sharing ad or search revenues.
In addition to the litigation and business solutions of private parties, government actors are sorting out various positions.
The U.S. Copyright Office has issued a report expressing views on infringement and fair use issues, a Congressional hearing called generative AI training “the largest intellectual property theft in American history,” and, in the last month, an executive order called for both a “minimally burdensome national standard” and the promise that “copyrights are respected.”
All of these threads contribute to a larger narrative: writ large, copyright law is reactive to technological changes.
Originally based on restricting duplication of printed materials, copyright has adapted to deal with the invention of broadcasting, personal copying devices, adoption of the internet and the rise of international trade and cultural exchanges.
Changes have occurred through both legislative action and judicial interpretation, but there have been growing pains while, as you might expect, technology and business move faster than the law.
Generative AI is another one of these inflection points, and we are in the middle of the awkward developmental phase. The name of the doctrine at issue—fair use—implies that an aspect of morality or justice is at play, but generative AI is still just too new to have established a normative consensus on how it should interface with copyrighted content.
In any event, AI is here and must be dealt with under the law we have today. Will 2026 see comprehensive solutions to the fair use problems with AI systems? Probably not, but 2025 shows us that some groups are finding enough of a solution to get by.•
__________
Blake Hartz is a partner at Woodard, Emhardt, Henry, Reeves & Wagner LLP. Opinions expressed are those of the author.
Please enable JavaScript to view this content.