An Anthropic Deal Won’t Jeopardize the Affordable AI Future

The tech world is abuzz with Bartz v. Anthropic—the largest reported settlement in a copyright infringement lawsuit to date. This eye-popping $1.5 billion figure has sparked widespread speculation: Is this the end of affordable, innovative artificial intelligence? Will the cost of licensing and compliance make it impossible for new platforms to emerge or for existing ones to remain accessible? The answer to these questions is almost certainly “no.”

Judge William Alsup, who oversaw and paused Anthropic’s proposed settlement, warned that class lawyers are striking a deal behind the scenes that will be “forced down the throat of authors.” Alsup later said approval of the settlement was postponed pending submission of further clarifying information.

The law and the marketplace will find a way to permit and nurture further innovation in AI, as it has done since the era of the printing press.

The settlement was about content acquisition, not training AI models. In June 2025, Alsup of the US District Court for the Northern District of California, held that training Anthropic’s AI model with copyrighted works is fair use. However, he declined to grant summary judgment to Anthropic for other actions. Relevant to this decision, Anthropic gathered texts to train its AI model in two ways.

First, it purchased hard copies of books, took them apart, scanned them, uploaded them, and converted them to readable text. This process helped Anthropic use the gathered texts to train its AI model.

Second, Anthropic illegally downloaded texts from “shadow libraries,” an act that Alsup deemed “piracy.” Anthropic then used those texts to train its AI.

For these methods, Alsup held that the copying of the texts for training the AI model was transformative and fair use. However, he considered the upstream method of how Anthropic obtained the texts as a separate use, to be judged on its own merits.

The scanning (i.e., copying) of the hard copies that Anthropic obtained legally was determined to be fair use, with the transformative purpose of shrinking the space needed to store the texts. However, according to the decision, downloading pirated texts had no transformative purpose. Anthropic argued that it downloaded the pirated texts for the purpose of creating a “central” library of all texts, but the court rejected the theory that this purpose was transformative, holding that it didn’t bring the use into the fair use exception.

Alsup granted Anthropic summary judgment of no infringement for the copying that was part of the actual training of the AI model and for the scanning of the hard copy books. However, he didn’t grant summary judgment for the act of downloading pirated books. It was plaintiffs who were affected by this upstream “piracy” that defines the class that settled its lawsuit with Anthropic.

Alsup’s decision and subsequent settlement drew a very thin line between the different acts that all had the same ultimate purpose: to train an AI model. There is case law to support this decision, holding that different acts, when taken together have a single fair use purpose, must be considered individually. There also is case law that would undermine this decision, holding that only the ultimate purpose of a use is to be considered in the fair use analysis.

Anthropic’s decision to settle this matter deprived us, for now, of a higher-court decision regarding this “individual act” versus “ultimate purpose” analysis, and as to whether the ultimate purpose of training an AI model is fair use. These issues are almost certainly going to be heard by circuit courts and, quite possibly, the US Supreme Court.

Copyright law evolved to support technological innovation. Despite the current uncertainty, history suggests that neither the law nor the marketplace will allow innovation to grind to a halt. Consider these examples:

• Digital Recording. The Supreme Court in Sony Corp. v. Universal City Studios (1984) held that recording TV shows for later viewing (time-shifting) was fair use, paving the way for VCRs, DVDs, and digital recording.
• Internet. Congress enacted the Digital Millenium Copyright Act in 1998, which provides safe harbor protection to websites such that they can’t be responsible for third-party copyright infringement, enabling the growth of the modern internet.
• Music Streaming. After courts shut down peer-to-peer sharing sites such as Napster, the market responded with legal streaming services such as Spotify and Pandora in the early 2000s.
• Online Search. In the early 2000s, the Ninth Circuit held that caching and low-resolution reproduction for search purposes was non-infringing, supporting the development of online engines. (Kelly v. Arriba)

Rapid and innovative development of AI will continue. What does the Anthropic settlement mean for the future of AI? Will AI become too expensive to develop or use? The answer, based on both the specifics of this case and the broader history of copyright and technology, is likely “no.”

The few district courts that have opined on copyright infringement and training AI data have all agreed in principle that training an AI model on unlicensed works may be fair use in some cases. The Anthropic settlement addresses only the act of acquiring data through piracy, not the act of training AI on lawfully obtained data.

Following the Anthropic decision and settlement, AI companies may amend the ways that they acquire materials for training their models, but the actual training may be protected as fair use. As with digital recording, the internet, and streaming music, the market and the law will evolve to enable practical and affordable ways to acquire content legally so that AI technology continues to develop. While the Anthropic settlement is a watershed moment, it shouldn’t be the end of affordable, innovative AI.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.

Author Information

Sarah Fink is of counsel in Wilson Elser’s intellectual property & technology practice.

Write for Us: Author Guidelines