Federal Courts Split on AI Training Copyright Fair Use in 2025
Three major 2025 federal court rulings reveal a fractured legal landscape regarding whether AI training on copyrighted materials constitutes fair use. While some judges find the practice highly transformative, others warn of significant market harm.
Three pivotal federal court decisions in 2025 demonstrate that United States fair-use doctrine is not moving in a single direction regarding artificial intelligence training on copyrighted works. These rulings show that the legal boundaries remain highly fractured, and it takes time for appellate courts to establish a unified approach to this complex issue.
In Bartz v. Anthropic, Judge William Alsup rules that training large language models is "spectacularly" transformative, comparing the process to human reading and learning. However, he draws a strict line against acquiring pirated books from shadow libraries, emphasizing that a transformative end use does not excuse the initial theft of copyrighted materials, a decision that leads to a $1.5 billion settlement.
Shortly after the Anthropic ruling, Judge Vince Chhabria tackles similar issues in Kadrey v. Meta by also favoring fair use for AI training but attaching a strong warning label regarding market harm. Together, these conflicting lower court opinions highlight the ongoing uncertainty for AI developers and the urgent need for higher courts to provide clear, consistent guidance.