On June 23 and 25, federal judges in the Northern District of California ruled in Bartz v. Anthropic and Kadrey v. Meta that using copyrighted books to train LLMs is transformative fair use, holding that training LLMs is a novel use distinct from the original publication purpose. The rulings diverge sharply on whether market harm must be proven empirically: Judge Alsup found no relevant market harm; Judge Chhabria required affirmative evidence from plaintiffs, noting that lack of proof defeated the authors' claims. Both judges emphasized that empirical evidence of market dilution or competitive harm could reverse the outcome. These decisions eliminate a major source of training-data IP liability for the two largest model vendors but leave open substantial exposure for copying and indefinite retention of pirated copies. For AI vendors and financial institutions holding equity in hyperscalers, the rulings materially reduce one quantifiable balance-sheet cost category but signal that future litigation risk turns on demonstrable market competition rather than copying alone.
275 other tracked developments touching the same claims.