With a deep understanding of your business alongside clear and honest communication, we help clients face challenges fearlessly.
Learn more about our services and how we help clients.
July 17, 2025
The artificial intelligence industry received a mixed verdict in a landmark AI copyright law case that could reshape how AI companies source, store, and use training data. In Bartz v. Anthropic, Judge William Alsup of the Northern District of California delivered a nuanced ruling that both validates AI training practices and warns against reckless data acquisition and storage—a decision with significant implications for business leaders navigating AI copyright law compliance.
The Case That Split the Difference
Writers and journalists Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson alleged that Anthropic had built its multibillion-dollar Claude AI system by illegally using pirated versions of their copyrighted works. The authors claimed Anthropic “stole hundreds of thousands of copyrighted books” to train its large language model, challenging the very foundation of how AI companies develop their systems.
On June 23, 2025, Judge Alsup handed down a split decision that gave both sides partial victories while establishing critical precedents for the AI industry. The court ruled that training AI models on lawfully purchased books constitutes fair use—calling Athropic’s use “exceedingly transformative” and comparing it to human learning. However, the judge also found that Anthropic’s practice of downloading over 7 million pirated books and maintaining them in a “central library” crossed the line into copyright infringement.
A Judicial Endorsement of AI Training—With Critical Caveats
The court’s fair use analysis provides some relief for AI companies that have been operating under legal uncertainty. Judge Alsup explicitly endorsed the transformative nature of AI training, stating: “Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them—but to turn a hard corner and create something different.” This reasoning aligns with broader legal principles that protect transformative uses of copyrighted material.
However, the court’s ruling comes with a critical warning: how you acquire and store your training data may have legal implications. Anthropic’s error—downloading millions of pirated books “just in case”—undermined its otherwise strong fair use defense and exposed the company to significant liability.
The Bartz v. Anthropic case isn’t over—Judge Alsup has ordered a December trial to determine damages, with potential statutory penalties of up to $150,000 per infringed work.
The Four-Factor Fair Use Analysis
Judge Alsup’s decision provides valuable guidance on how courts will evaluate AI training under traditional fair use doctrine:
Purpose and Character of Use: The court strongly favored AI companies here, finding that LLM training builds generalized knowledge rather than reproducing original works.
Nature of the Work: While the court acknowledged that creative works like novels typically receive stronger copyright protection, this factor weighed only modestly against fair use given the transformative nature of AI training.
Amount Used: Even though AI systems typically ingest entire works, the court found this necessary for models to learn language patterns and meaning, resulting in a neutral finding on this factor.
Market Impact: The court found no concrete evidence that AI training harmed the market for the original books, noting that AI outputs don’t serve as replacement products. However, the ruling warns that substantial market displacement could undermine fair use protections in future cases.
Implications for Business Strategy
Companies can proceed with greater confidence when using lawfully acquired content for AI training. The judicial recognition of AI training as fair use provides a legal foundation for business strategies that seemed uncertain under traditional AI copyright law frameworks just months ago.
However, the court’s emphasis on lawful acquisition means businesses must implement robust data governance practices and maintain clear audit trails showing how training content was sourced—a fundamental requirement for AI copyright law compliance.
Lessons from the Broader Legal Landscape
The Bartz v. Anthropic decision aligns with other recent AI copyright rulings while distinguishing itself through its nuanced approach to data acquisition. The parallel ruling in Kadrey v. Meta similarly affirmed the fairness of AI training, while the Thomson Reuters v. Ross Intelligence decision—which we previously analyzed—rejected fair use arguments around legal research datasets.
These cases signal that courts are willing to protect transformative AI applications while maintaining strict AI copyright law standards for data acquisition and use.
Practical Guidance for Business Leaders
Prioritize Licensed and Lawfully Acquired Data: The court’s split decision makes clear that scanning legally purchased books fits within fair use, but downloading pirated content remains risky. Businesses should invest in legitimate data sources and comprehensive licensing agreements.
Document Your Data Pipeline: Maintain detailed records showing how each element of your training data was acquired and why it was necessary for your AI system’s development. This documentation could prove crucial in defending against future copyright claims.
Avoid Building “Central Libraries” of Questionable Content: Businesses should implement policies that isolate and purge content of questionable origin.
Prepare for Market Impact Analysis: While the court found no current market harm, future cases may require businesses to demonstrate that their AI outputs don’t compete directly with copyrighted works. It will be interesting to see how market impact and monetization come into play in the ongoing Disney and NBCUniveral v. Midjourney case.
The Road Ahead
The Bartz v. Anthropic decision provides important guidance for navigating the balance between the potential of AI technology and respect for intellectual property rights.
More broadly, this decision represents an evolution of AI copyright law. Courts are moving beyond blanket approvals or rejections of AI training toward nuanced analyses that consider both the transformative potential of AI and the legitimate rights of copyright holders under evolving AI copyright law precedents.
For businesses operating in this space, the AI copyright law message is clear: AI training can proceed under fair use protection, but only with careful attention to data sourcing and storage practices.
At BoyarMiller, we’re committed to helping clients understand and navigate these complex legal developments. As your trusted advisors, we’ll continue monitoring these cases and providing the practical guidance you need to leverage AI technologies while managing legal risks effectively.
With a deep understanding of your business alongside clear and honest communication, we help clients face challenges fearlessly. Learn more about our services and how we help clients.