While artificial intelligence ("AI") as we know it today has only been mainstream for several years due to the rapid advancement of generative AI platforms like ChatGPT, AI's warp speed development and low barrier to entry, both for everyday consumers and businesses, have created a legal landscape reminiscent of the dot-com boom of the 1990s where the legal system was slow to keep pace with fast-moving technologies. Simply put, the legal system has struggled to apply 20th Century legal principles governing copyright law to AI while the leading platforms have been allowed to grow seemingly unchecked.
However, two recent summary judgment decisions by U.S. Federal District Court judges in the Northern District of California, one by Judge William Alsup in Bartz et al. v. Anthropic PBC (June 23, 2025), and another by Judge Vince Chhabria in Richard Kadrey et al. v. Meta Platforms, Inc. (June 25, 2025), are beginning to establish a framework for how federal courts will address the intersection of copyright, AI training, and the boundaries of fair use. Together, these rulings offer valuable insight into how courts may balance the protection of creative works with the need to foster rapid AI innovation.
This alert breaks down these pivotal decisions, highlights other key cases and regulatory developments, and outlines what it all means for clients across industries. While innovation will always outpace regulation, understanding the evolving legal landscape is essential for managing risk and seizing opportunity.
What the Courts Held
Anthropic
A group of authors brought a class action against Anthropic PBC, alleging that the company infringed their copyrights by using their books, some lawfully purchased, others pirated, to train its large language models (LLMs) and create a digital library for further training. Judge William Alsup ruled that Anthropic's use of lawfully acquired, digitized books for AI training constituted fair use, emphasizing the "spectacularly" transformative nature of this process. The court analogized AI training to a person internalizing knowledge from books, noting that such learning does not require ongoing royalties or licenses.
Importantly, the court found that the outputs of Anthropic's Claude model were not alleged to be direct reproductions or close substitutes for the original works, and thus did not infringe the authors' copyrights. However, the court drew a sharp line at Anthropic's use of over seven million pirated books, finding that the unauthorized acquisition, storage, and use of these works was not protected by fair use and could give rise to liability at trial.
Meta
In a parallel case, Meta's use of copyrighted books to train its LLaMA model was also challenged. Judge Vince Chhabria ruled in favor of Meta, finding that its use of the books, even though unauthorized, qualified as fair use because the works were used solely as raw material to teach language patterns, not to reproduce or compete with the originals. The court emphasized that the plaintiffs failed to show that LLaMA's outputs substituted for or harmed the market for the original works. Notably, the court also held that DMCA takedown claims failed because Meta did not distribute or publicly display the works; the books were used internally for training purposes only.
Broader Legal Landscape
- Thomson Reuters v. ROSS Intelligence (D. Del. 2025): In contrast to Anthropic and Meta, the Delaware court found that ROSS's use of Westlaw's proprietary headnotes to train an AI legal research tool was not fair use. The court emphasized that ROSS's use was not transformative and directly competed with Westlaw's core market, undermining its primary revenue stream and demonstrating meaningful market harm.
- Google v. Oracle (U.S. Supreme Court 2021): The Supreme Court held that Google's copying of 11,500 lines of code from Oracle's Java SE program for use in Android was fair use, underscoring the importance of "transformative" use in fair use analysis. The ruling went on to state that "Google's copying of the Java SE API, which included only those lines of code that were needed to allow programmers to put their accrued talents to work in a new and transformative program, was a fair use of that material as a matter of law." This principle was cited by Judge Alsup in the Anthropic decision as central to evaluating AI training on copyrighted works.
- U.S. Copyright Office AI Report (2025): The Copyright Office's recent reports reiterate that fair use is a fact-specific inquiry, requiring careful consideration of purpose, nature, amount, and market effect. The Office cautions that acts such as downloading, storing, and training on copyrighted works often implicate exclusive rights, and there is no categorical safe harbor for AI training under fair use. Only courts will ultimately define these boundaries, making legal guidance essential.
What This Means for Clients: Managing the Content and Data Lifecycle
- Know Your Sources and Uses: Understand exactly where your content and data come from and where they may end up. Unauthorized scraping or downloading remains a significant infringement risk, as fair use defenses typically apply only after acquisition.
- Maintain Robust Documentation: Keep thorough records of how you acquire, use, and protect your content and data. Clear documentation not only supports compliance but also strengthens your position in future disputes.
- License Proactively and Precisely: Not all licenses are created equal. Ensure your agreements explicitly address AI training and data and content use rights and include provisions that protect your interests as the law evolves. With courts still defining the market for AI training rights, proactive licensing can strengthen your position.
- Educate and Enforce Internally: Make sure your teams and partners understand the boundaries of fair use and the risks of unauthorized data or content acquisition and usage. Regular training and clear internal policies are key.
- Audit and Assess Regularly: Conduct periodic audits of your data and content pipelines. The risks of using pirated or unlicensed material are both legal and reputational, as recent cases have shown.
- Stay Informed and Prepare for Change: Monitor developments from the courts, the Copyright Office, and Congress. Staying ahead of regulatory and legal changes will help you anticipate risks and seize new opportunities. Regularly review and update your contracts, policies, and risk assessments to stay compliant and competitive.
Conclusion
The Anthropic and Meta rulings mark a pivotal moment in U.S. copyright law as applied to AI, but they will not be the last word. Divergent court decisions (such as Thomson Reuters v. ROSS), ongoing legislative debates, and evolving regulatory guidance mean that best practices and legal risks will continue to shift. Clients should act now to audit practices, strengthen compliance, and position themselves for a dynamic legal environment.
We regularly counsel clients ranging from AI developers and software providers to fashion companies, media organizations, and other creative and technology-driven businesses. Our team is experienced in navigating the complex and evolving landscape of copyright, licensing, and AI, and we are ready to assist with tailored advice to address your specific needs and challenges.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
