The frequency of lawsuits involving the development and deployment of AI technologies is increasing by the day. Recent lawsuits seeking to hold companies directly and secondarily liable for “joint enterprises” based on use (or alleged misuse) of copyrighted works for training AI models serve as important reminders about the protections that intellectual property (IP) insurance can offer to cover the risks associated with copyright infringement claims.

Recently, a California federal district court ruled that it was “fair use” for an AI software company to use copyrighted books to train its large language models (LLMs). However, the court also found the company’s unauthorized possession of over seven million pirated books that it downloaded from the internet (apparently for free) amounted to copyright infringement independent from whether the books were ultimately used to train the LLMs. In contrast, where the company purchased books before scanning them into digital files, the use was a permissible “fair use.”

The court’s order in Bartz et al. v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal. June 23, 2025), highlights the nuanced permissible use of copyrighted training data and underscores why policyholders engaged in the use of copyrighted material should acquire and maintain robust IP insurance that will reliably respond to claims of alleged infringement.  

Factual Background

The company, Anthropic PBC, is an AI software company whose main offering is its AI software, Claude. Anthropic acquired copyrighted works by purchase and internet pirating to create a digital “central library,” some of which was used to train Claude. Anthropic planned to maintain the central library forever, regardless of whether a book was eventually used to train its LLMs.

Three authors brought a putative class action against Anthropic for its unauthorized copying of the authors’ books. The plaintiffs alleged Anthropic had infringed their copyrights by pirating copies of their books for its libraries and reproducing them to train LLMs. Anthropic maintained that it copied the authors’ works solely for the purpose of training its LLMs. The authors argued, however, that the works were used for at least two distinct purposes: to create a comprehensive digital library of potentially valuable material and to train various LLMs by selecting and refining subsets of that library over time, favoring works with stronger organization and expressive quality.

Anthropic moved for summary judgment on whether its use of the copyrighted materials for its library and LLM training constituted a “fair use.” The court granted summary judgment to Anthropic as to its training of LLMs, finding that such use was a fair use of the printed and digitized versions of the books. However, the court denied summary judgment to Anthropic as to whether the pirated copies downloaded from the internet were being fairly used. The court further found that the use of pirated copies to create Anthropic’s central library and resulting damages presented an issue for trial. 

The Court’s Fair Use Analysis  

The court distinguished between Anthropic’s different uses of the copyrighted books--namely, to create a digital library, and to train its LLMs. The court also distinguished between the pirated copies and the legitimately purchased physical copies of books that were converted into digital form in its fair use analysis.

Below, we focus on those portions of the court’s fair use analysis that could result in liability for copyright infringement claims. Policyholders facing similar claims may be able to use IP insurance to help cover the costs of their legal defense, as well as any settlement or judgments for which they are found liable.

Section 107 of the Copyright Act requires courts to consider four factors in evaluating a question of “fair use.” Fair use is a legal doctrine that allows the unlicensed use of copyright-protected works in certain situations. Different types of uses of copyrighted works—such as for criticism, commentary, news reporting, and teaching for example—may qualify as fair uses under the Act.

The four factors contemplated to determine fair use are: (1) the purpose and character of the use; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and (4) the effect of the use upon the potential market for or value of the copyrighted work. We discuss each below.

1. Purpose and Character of the Use

The first fair use factor focuses on the purpose and character of the use. Under this factor, courts consider whether a work is transformative, i.e., it adds new expression, meaning, or message to the original work, and whether copyrighted material is used for commercial purposes. Transformative works (e.g., mashup songs or using thumbnail images in search results) and non-commercial uses (e.g., academic research or personal use) of copyrighted work are more likely to be considered fair use.

In Bartz, the court held that building and indefinitely retaining a comprehensive virtual library of pirated copies of the books—regardless of whether they were ultimately used—was not a fair use because Anthropic did not pay for the digital copies of the books it copied from pirated websites and retained them regardless of whether the copies would become useful, which was not a transformative use of the books. In further support of its finding, the court noted that the pirated copies of books were not immediately transformed into a significantly altered form, and were kept after they were used to train LLMs. Further, Anthropic did not instill internal controls limiting access to and use of the copies.

2. Nature of the Copyrighted Work

The second fair use factor rests on the nature of the copyrighted work. This factor examines the degree to which the work that was used relates to the Copyright Act’s purpose of encouraging creative expression. The use of a more creative work like a book is less likely to support a claim for fair use than a factual work such as a technical report or article. Here, the court acknowledged that the works at issue contained expressive elements, and the court accepted the authors’ evidence that their books were chosen for their expressive qualities by Anthropic in building its library and training its LLMs. The court found the second factor weighing against fair use for both the printed copies and the pirated digital copies of the works irrespective of Anthropic’s use.

3. Amount and Substantiality of the Portion Used

The third fair use factor focuses on the amount and substantiality of the portion of the copyrighted work used by the copyist. The amount of copying for this factor is considered first against the work itself, and then against the proposed transformative purpose. Generally, if a large portion of a copyrighted work is used, fair use is less likely to be found by courts. Sometimes, the use of entire works is considered fair in certain circumstances.

For the pirated copies of the books, the court found Anthropic lacked the right to hold copies of these books at all. The court noted Anthropic’s retention of the books, even if some of those books were not going to be used to train its LLMs, demonstrated there were other further uses, which weighed against fair use, as almost any copying of the books that were acquired on the chance some of them might be useful for training LLMs and other uses would have been too much, according to the court.

4. Effect on the Market

The final factor for fair use is the effect of the copyists’ use on the potential market for or value of the copyrighted work. In assessing this factor, courts look at whether the copyists’ use is hurting the current market for the original work by displacing sales of the original work and/or whether the copyists’ use could cause substantial harm if it were to be far-reaching. This factor weighs against fair use when a copyist makes copies available that displace demand for copies of works a copyright owner already makes available or easily could.

The court found the pirated copies of books used to form Anthropic’s internal library is akin to permitting Anthropic to steal a work it could otherwise purchase as long as it intended to use those copies for an allegedly transformative use such as training its LLMs. The court found that this would destroy the publishing market if this were the case, and that the market harm was more significant for the copies used to generate Anthropic’s digital library, weighing against fair use.

Discussion

As the decision in Bartz shows, the risk of a claim for IP infringement is becoming increasingly common, particularly with the rise of emerging technologies such as AI. The good news is that IP coverage may be available for such risks. IP coverage can offer protection for claims related to copyright or trademark infringement.

Risks like copyright or trademark infringement and the existence of IP insurance are not novel concepts. For decades, court dockets across the country have been filled with cases involving actual or alleged infringement of creative works in various forms, with policyholders often seeking coverage for these disputes under their IP insurance policies.

Common types of IP coverage include infringement defense and abatement enforcement coverage. Infringement defense helps cover a company’s costs if a lawsuit is brought against it for IP infringement, such as the Bartz lawsuit against Anthropic. Abatement enforcement coverage helps pay the costs for a business to pursue legal action against another person or business alleged to infringe upon a company’s IP. Some IP coverage forms also offer protections for patent infringement lawsuits.

Traditionally, companies that may benefit from IP coverage include:

• Tech Startups. Companies that develop their own software, or other technology products, are vulnerable to third-party infringement claims and should consider making IP coverage part of their current insurance program. As an example, Anthropic, as part of its development of Claude, makes its own technology products and software and could benefit from IP coverage.
• Entertainment and Media Companies. Companies in the film, music, and other industries involving the use and development of creative works would benefit from the protections that IP coverage can afford in the event they receive an infringement claim or need to pursue a claim against an infringer.
• Consumer Products Manufacturers and Designers. Companies that design and manufacture retail and consumer products such as food and beverage products, cosmetics products, and others may benefit from the protections of IP insurance to pursue claims against competitors and counterfeiters. 

Takeaways for Policyholders

IP insurance protects companies in the event claims of infringement are asserted against them. IP coverage can also help companies cover the costs incurred in the course of pursuing claims against third-parties for misappropriation of their IP rights. Such costs can include legal fees, settlements, and adverse judgments.

Policyholders should pay close attention to the language of their current liability policies, as the availability and scope of potential IP coverage varies widely. Traditional policies, like commercial general liability and media liability, may exclude certain types of IP-related exposures altogether. But even insurance products tailored specifically for IP exposures can vary in scope.

Standalone infringement liability coverage may be available for infringement claims by competitors and other IP holders. Policyholders whose work involves creating IP or leveraging third-party IP also should assess their current insurance programs to evaluate the availability and scope of coverage for IP-related claims. Even if a policyholder does not own or create IP, any business engaged in the manufacturing, distribution, wholesale, retail, or service of products can be accused of IP infringement. Experienced coverage counsel and other risk professionals can help ensure that policyholders are not left with unintended risk exposures in the event of an IP-related dispute.