The question of whether U.S. copyright law permits AI companies to train models on copyrighted works has moved from academic debate to courtroom reality. In 2025, three federal district court decisions began to outline what counts as fair use in this context, according to a single source. These rulings are early attempts to apply existing copyright principles to a technology that did not exist when the law was written.
The decisions do not settle the issue, but they provide the first concrete judicial guidance on how factors like transformation, market harm, and the nature of the training data might be weighed. The source notes that the rulings "began to sketch the boundaries," implying that the courts are still working from a blank canvas. Each case likely turned on its specific facts, leaving room for future courts to draw different lines.
Because the source is the only one available, there is no competing view to compare. Still, the takeaway is clear: the legal landscape for AI training is being built case by case, and 2025 marked a turning point where theory gave way to precedent. Companies and creators will be watching closely as appeals and new lawsuits refine these initial rulings.