A U.S. Judge has ruled that a company’s use of copyrighted material to train its generative AI is considered fair use in law. It’s a potentially landmark decision that could have huge consequences for both generative AI and content creators. However, it’s not a straightforward victory for the AI company as it still faces charges of piracy.
What was the legal case?
U.S District Judge William Alsup of the Northern District of California gave the ruling earlier this week. The legal decision came after three authors brought a lawsuit against the AI company Anthropic. One of the authors was best-selling mystery thriller writer Andrea Bartz, who wrote titles including “We Were Never Here” and “The Last Ferry Out.”
The other plaintiffs were the non-fiction writers Charles Graeber and Kirk Wallace Johnson. All of the authors claimed that Anthropic had stolen their work to train the company’s Claude generative AI model.
What had Anthropic done?
In order to train its Claude generative AI model, Anthropic had purchased physical copies of printed books. The company then stripped the pages from the books and scanned them into PDF files. However, Anthropic also sourced books from illegal, pirated libraries.
The legal ruling stated that the company “could have purchased books, but it preferred to steal them to avoid ‘legal/practice/business slog,’ as cofounder and chief executive officer Dario Amodei put it.”
In total, Anthropic pirated over seven million copies of books. These books included copies of at least two works for each of the three Plaintiff authors.
What did the Judge say?
Judge Alsup ruled that Anthropic’s use of copyrighted material to train its Large Language Model (LLM) generative AI was permitted as fair use. He stated, “Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works, not to race ahead and replicate or supplant them — but to turn a hard corner and create something different.” Judge Alsup went on to clarify that “the use of the books at issue to train Claude and its precursors was exceedingly transformative and was a fair use under Section 107 of the Copyright Act.”
What next?
Despite Judge Alsup’s ruling on fair use, the outcome of the case wasn’t all good news for Anthropic. The Judge went on to determine that Anthropic’s use of pirated material broke the law. He said that Anthropic had saved pirated copies of their books as part of a “central library of all the books in the world.” As such, the company had violated the authors’ rights, and Judge Alsup will hold a second trial later this year to determine if damages are due. Under U.S. copyright law, Anthropic could have to pay damages of up to $150,000 for each pirated work.
What we think
Judge Alsup’s ruling that training an AI on copyrighted material could have huge implications for the future development of generative AI. The Judge likened the training to a human reading all the modern-day classics before going on to write their own book. It’s a blow to content creators who wanted to restrict the use of their copyrighted works. At the same time, it’s a potential green light for AI companies to use whatever material they want to train their generative AI models. However, the one positive for creators is that the AI companies must legally purchase and own the copies of the works they use. By failing to do this and using pirated material, Anthropic is potentially facing a big bill for damages.
