Source: IPWatchdog

IPWatchdog provides a comprehensive review of the 2025 legal landscape where U.S. copyright law and artificial intelligence have collided in three pivotal district court decisions. The year 2025 marked the end of the theoretical era, as federal judges began to define the boundaries of "fair use" for AI training. In Bartz v. Anthropic, the court favored the developer, finding the training process to be highly transformative and unlikely to displace the market demand for authors' original works. This decision highlighted the importance of factor four in fair use analysis: the effect on the potential market value of the copyrighted work.

Conversely, the Kadrey v. Meta case offered a more cautious perspective, where the judge granted partial summary judgment for the developer but issued a warning regarding the potential for market harm, especially when models are trained on unlicensed "shadow libraries." The third critical case, Thomson Reuters v. ROSS Intelligence, saw the court take a dim view of AI training that directly substitutes for a paid product. In this instance, the court held that using copyrighted headnotes to build a competing legal search platform did not constitute fair use, as it effectively "cannibalized" the original owner’s market.

The article emphasizes that these decisions show the U.S. fair-use doctrine is currently moving in multiple directions, with a unified approach unlikely until appellate courts weigh in during 2026. The U.S. Copyright Office has also taken a preliminary stand, suggesting that the speed and scale of AI generation pose a serious risk of diluting markets for original human creators. For AI companies, the key takeaway from 2025 is the necessity of rigorous documentation and legal analysis of training methods. As the legal target continues to move, early attention to these precedents remains the best strategy for managing uncertainty and mitigating massive financial exposure.