“From the infancy of copyright protection, some opportunity for fair use of copyrighted materials has been thought necessary to fulfill copyright’s very purpose, ‘[t]o promote the Progress of Science and useful Arts….’” Campbell v. Acuff-Rose, 510 U.S. 569, 575 (1994). The question, however, is how much—particularly given the use of books and other materials to train AI models. LLMs, such as Anthropic’s Claude, have used these materials to train their models. In training, they use not just quotes or summaries, but the entirety of works.[1]
This use has inevitably resulted in litigation testing the bounds of what constitutes fair use (see Copyright Office overview HERE). The most prominent case to emerge was Bartz v. Anthropic, in which a group of authors sued Anthropic over alleged copyright infringement. This case has tentatively settled, with massive implications for the future of copyrighted materials and AI. This post breaks down the settlement, its implications, and some unresolved legal questions.
Settlement
On Friday, September 5, 2025, a $1.5 billion settlement was announced between Anthropic and authors of 500,000 works for roughly $3,000 per copyrighted work. This settlement was seen as a major win for authors. Some observers noted that it was akin to rulings against Napster in the early 2000s. The settlement, however, requires judicial approval—which is now up in the air. On Monday, Judge William Alsup skewered the settlement, noting that it is full of pitfalls.
Judge Alsup issued a mixed ruling on the case in June. He dismissed some of the authors’ claims, finding that use of books to train AI constituted fair use. Judge Alsup, however, did not dismiss claims related to the method of acquiring the books, which were set to go to trial before the settlement. If the settlement falls apart, these questions could be litigated later this year or next year.
What does this mean?
The fate of the Anthropic settlement still hangs in the balance. The case, however, has massive implications for companies looking to deploy AI. Here are a few to consider:
- Get Proper Access. Courts have thus far granted wide latitude when works are acquired legally. Training AI models is considered fair use of a copyrighted work. However, the work can’t be pirated or acquired without a valid license. Companies that are looking to build or use LLMs—or otherwise utilize AI—must have a strategy for legally licensing works that they want to use for training purposes.
- Fair Use Standards Still Apply. While use of copyrighted work to train models is novel, the four-factor Campbell v. Acuff-Rose framework still applies. Companies in the AI space should not reproduce substantial portions of copyrighted works. When deciding whether a use of a copyrighted work qualifies for fair use, judges consider—and will continue to consider —four factors:
- the purpose and character of use
- the nature of the copyrighted work
- the amount and substantiality of the portion taken, and
- the effect of the use upon the potential market.
- Monitor Developments. This area of law is evolving rapidly. The U.S. Supreme Court may eventually weigh in. Businesses should not assume that the Anthropic settlement is the last word on the subject. It will be crucial to stay informed regarding the latest developments in AI and copyright.
- Expect Licensing Markets to Emerge. Even if training is often deemed fair use, settlements like Anthropic’s show that there is enormous pressure to compensate rights holders. Napster showed the potential of digital distribution but collided with copyright law. The legal fallout accelerated the creation of structured licensing markets (iTunes first with per-track sales, later Spotify with subscription streaming). These platforms created legal, monetized channels. Expect something similar with training-data licensing.
AI is shaking up the world of copyright and fair use. The line between innovation and infringement is being redrawn in real time.
Disclaimer: This blog is for informational purposes only and does not constitute legal advice. Reading or interacting with this content does not create an attorney–client relationship. You should consult a qualified attorney for advice regarding your specific situation. Mehaffy, PLLC disclaims all liability for actions taken or not taken based on this blog.
[1] As the U.S. Patent and Trademark Office has observed, LLM “training” “almost by definition involve[s] the reproduction of entire works or substantial portions thereof. 18 U.S. Patent & Trademark Office, Public Views on Artificial Intelligence and Intellectual Property Policy 29 (2020), available at: https://www.uspto.gov/sites/default/files/documents/USPTO_AIReport_2020-10-07.pdf (last accessed Sep. 9, 2025).