Buy-side investors typically require the target company to represent that it has not infringed on the copyrights or other intellectual property of any third party. Typical language might read:
“The operation of the Company as presently conducted does not infringe, dilute, misappropriate or otherwise violate, nor has infringed, diluted, misappropriated or otherwise violated, the Intellectual Property rights of any third person….”
This representation can be more complicated for the growing number of generative AI businesses which train their large language models (LLMs) on copyrighted materials. Such training includes copying entire books and other copyrighted materials. These businesses must understand, well in advance of a transaction, whether their use constitutes permissible “fair use” or infringement under the Copyright Act.
The Copyright Act lists four factors to be considered in determining whether a given use is fair, summarized as follows: (1) the purpose and character of the use; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used; and (4) the effect of the use upon the potential market for or value of the copyrighted work.
If the use of the copyrighted work has a further purpose or different character from the original, and does not substitute for the original use, it may be deemed to be “transformative,” a characteristic weighing in favor of fair use. Parody is a classic example, allowing creators to commercially exploit content despite containing elements of the original work.
Three recent court decisions highlight the complex and evolving framework for applying these Fair Use factors to the unauthorized use of copyrighted works for training AI models.
In Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc. (D. Del., February 11, 2025), the court found that the defendant’s use of plaintiff’s copyrighted legal headnotes in order to train a non-generative AI-powered legal search tool was not “transformative,” and was commercial and directly competitive, and therefore amounted to copyright infringement (i.e., not a fair use).
In Bartz v. Anthropic PBC (N.D. Calif., June 23, 2025) and Kadrey v. Meta Platforms, Inc. (N.D. Calif., June 25, 2025), the court held that using legitimately obtained, copyrighted books for training large language models (LLMs) (i.e., generative AI) was “transformative”, which favored a finding of fair use rather than infringement. In Kadrey, the Court discarded the third factor, “amount used” even though entire books were copied as not especially relevant given the “reasonably necessary” and “highly transformative” use. Also focusing on the fourth factor (market dilution), the Court noted that the plaintiffs failed to present empirical evidence of dilution. In both cases, the court distinguished use of purchased books (fair use) from pirated books (not fair use).
Overall, these narrow opinions illustrate that training of non-generative AI which competes directly with the copyright owner is likely to be infringement (Thomson Reuters), whereas generative AI training (e.g., training of LLMs) is often transformative and permissible if the copyrighted content is lawfully sourced (Bartz and Kadrey). They do not address the various other types of generative AI, such as producing similar works of copyrighted art.
Going forward, Buyers and investors targeting AI owners will seek confirmation that the target’s use of copyrighted materials constitutes non-infringing, fair use. Buyers and investors will want to review the target’s IP licenses, data governance practices, and policies related to training data and AI development
In view of Bartz and Kadrey, AI owners should anticipate being asked to represent that their training data was obtained legally and in compliance with applicable laws. AI owners may try to negotiate knowledge qualifiers or other carveouts. Importantly, AI owners should be confident that their use of copyrighted works for training their LLMs has a different or further purpose than the original work, and does not harm the market for the original work. AI owners should consult with IP counsel for guidance in anticipation of making such representations.
