Should AI companies be liable for training models on copyrighted content without permission?

Context

AI companies train large language and generative models on vast datasets of text, images, and audio scraped from the internet—often including copyrighted books, articles, music, and artwork—without explicit permission or compensation to creators. The question of whether these companies should be held legally liable for this practice sits at the intersection of intellectual property law, innovation incentives, and creator rights, with major lawsuits from music labels, authors, and artists now testing the boundaries of fair use doctrine. The outcome will shape whether AI development requires licensing agreements with rights holders or whether training on copyrighted material remains permissible under existing law.