In a landmark decision, a federal judge has ruled in favor of Anthropic, determining that training its AI models on physical books that the company legally purchased does constitute fair use. This ruling marks a significant moment for the artificial intelligence sector, though it is specifically limited to the physical books acquired and digitized by Anthropic.
Judge William Alsup of the Northern District of California emphasized in his ruling that Anthropic will still need to face a separate trial regarding allegations of illegally copying “millions” of books from the internet. Furthermore, the ruling did not settle the question of potential copyright infringement by the outputs of AI models, an issue that continues to be litigated in other cases.
The lawsuit, initiated by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, alleged that Anthropic utilized pirated materials to train its Claude AI models. This decision is crucial as it may influence future judicial interpretations in other AI-related copyright cases.
The ruling also considered Anthropic’s methodology of purchasing print editions of books, removing their bindings, cutting the pages, and scanning them into a central digital library for AI training. The judge affirmed that digitizing a legally purchased book is fair use, and utilizing those digital versions for the training of a large language model (LLM) is transformative enough to also qualify for fair use protection.
“The authors’ complaint is akin to claiming that educating students to write well can lead to an influx of competing works,” Judge Alsup noted, highlighting that the Copyright Act aims to promote original creations instead of shielding authors from competitive pressures.
Despite the favorable outcomes for Anthropic, Judge Alsup also pointed out that the company’s practice of maintaining millions of pirated book copies in its library—regardless of their use in training—does not fall under fair use. “This order questions whether any accused infringer can adequately justify why they would download source copies from piracy sites rather than purchasing or legally accessing the materials elsewhere,” Alsup stated, emphasizing his skepticism.
The court will conduct a separate trial to address the issues surrounding the pirated content attributed to Anthropic, which will also ascertain damage amounts.
Anthropic spokesperson Jennifer Martinez expressed satisfaction with the ruling, stating, “We are pleased that the Court recognized that using works to train LLMs was transformative—spectacularly so. Aligning with copyright’s intent to foster creativity and scientific advancement, Anthropic’s LLMs were created not to simply replicate existing works but to innovate and produce something distinct.”
Update, June 24th: A statement from Anthropic has been added.
