As generative AI technology advances, the legal battle over the use of copyrighted materials to train these models is intensifying. In the first wave of lawsuits, courts are divided into approaches to fair use as a defense to claim copyright infringement. Other legal protection theories, including publicity and rights to unfair competition under state and federal law, have been rarely tested.
Fair use is a legal doctrine that allows for the restriction of copyrighted material without the need for permission from the rights holder. 17USC§107. The court will consider four main factors to determine whether the use is eligible for fair use. (1) Purpose and character of use (commercial or non-commercial purposes). (2) the nature of copyrighted work (factual or creative). (3) The amount and substantiality of the portion used. (4) Impact of market use.
In May 2025, Copyright Office released a third (pre-published) report providing ample support to both parties on whether AI constitutes copyright infringement or fair use in copyrighted works.
So far, in 2025, four cases directly analyzed the applicability of fair use defenses to AI model training. In most cases, each case put its own spin on the issue, confirming that the discussion on AI and intellectual property was only just beginning.
Lehrmanv. Lovo, Inc. , no. 1:24-CV-03770 (July 10, 2025)
The class action lawsuit filed in May 2024 by two audio actors from the Southern District of New York was one of the first actors to raise the issue [3]. Plaintiffs are Lovo, Inc. claims that they used replicas generated by voice AI without permission. Their 313 paragraph amended complaints result in a 17 count ranging from violations of New York law protecting rights to copyright infringement and various unfair competition and false advertising claims under the Lanaham law. Lovo moved to rejection, claiming that the plaintiffs failed to make a viable claim against the use of voices generated by AI as they are not copyright infringement and are not subject to relevant state law. The court held that claims for copyright infringement and violation of state law were well stated, but dismissed the claims under the Lanham Act. The court also allowed the actor to modify the allegations that Robo's AI training infringed copyright. Most critically for plaintiffs, class action lawsuits lovo I'll continue to live.
Bartz v. Human PBC, No. 24-CV-05417 (Nd Cal. June 23, 2025)
Humanity trained the AI model, Claude, using millions of digital books. I scanned many printed books, but I also downloaded other books from Pirate Library. The group of authors sues humanity for copyright infringement and argues that it is illegal to use them to copy books and train Claude. Humanity argued that its actions were fair use and sought an early summary judgment. The judge looked at three major activities individually and determined whether each was fair use.
- training: The judge determined that he trained an AI model on a legally acquired copyrighted book, constituted fair use and found its use to be highly transformative.
- Library copy: The judge also agreed that book purchases and digital books are transformative because “all printed materials purchased were copied to preserve storage space and enable searchability as digital copies.” However, the judge rejected a summary judgment regarding “it was made from a copy of the Central Library, but not used for training.”
- Pirated Library Copy: The judge rejected summary judgment because the use of these books was transformative due to training, but the creation and maintenance of a permanent, generic digital library of pirate works was not protected by fair use.
Kadreyv. MetaPlatforms, Inc. No. 23-CV-03417 (Nd Cal. June 25, 2025)
A few days later, another Northern District Judge in California admitted meta in a copyright lawsuit brought by 13 authors who alleged that their books were being used inappropriately to train Meta's llama AI models. The judge ruled that the plaintiffs' activities were found to be very transformative in training AI models, which were recognized as fair use under copyright law, and that the purposes such as summaries and content generation were different from the original work. The judge added that the plaintiffs did not provide sufficient evidence of harm. However, the judge emphasized that the court's decision was narrowly determined to focus on plaintiffs' inadequate arguments and that it did not establish the legality of the blanket due to AI training practices. He warned that more persuasive evidence in future cases could lead to different outcomes, and evidence of infringement or economic damage could be successful.
Thomson Reuters is coming in. Ctr. Gmbhv. Ross Intel. Inc.765 F. Supp. 3d 382 (D.del. 2025) (About appeals prior to 3rd circuit)
It's the opposite of the previous case Thomson Reuters is coming in. Ctr. Gmbhv. Ross Intel. Inc.the Delaware area has determined that fair use is an inapplicable defense when training AI models. Thomson Reuters alleged that Ross infringed copyright when he trained Ross' new AI legal research search engine using Thomson Reuters' Westlow's headnote. The court granted a partial summary judgment to Thomson Reuters on the claims of direct infringement, denying Ross' defenses, including fair use. When assessing Ross' fair use defense, the court found that Ross' use was “commercial” and not “transformative” and therefore in part not fair use. The decision is currently being appealed by Ross.
So what now?
The legal environment for artificial intelligence is still evolving, and it is not possible to predict results with any kind of accuracy. Some courts appear poised to accept AI model training as transformative, while others do not. As AI technology continues to advance, legal systems must adapt to address the unique challenges it presents. Meanwhile, companies and creators navigating this uncertain terrain should continue to inform legal developments and consider proactive measures to mitigate risk. One thing is clear as we await further ruling and potential legislative measures. The conversation about AI and existing intellectual property protection has just begun.
