AI Training on Copyrighted Books: A Complicated Legal Terrain

In a landmark decision, Judge William Alsup ordered Anthropic to pay a $1.5 billion settlement to a group of writers whose books were used to train the company’s large-language model. While the fine addressed Anthropic’s use of pirated copies from shadow libraries, the judge concluded that the act of training the model was lawful, likening it to a reader studying literature rather than copying it. A separate case involving Thomson Reuters and the research firm Ross Intelligence saw Judge Stephanos Bibas rule that Ross’s training on Reuters content was not “transformative” and therefore not protected by fair-use doctrine.

TechCrunch

Attorney Cathy Gellis, who specializes in intellectual-property law, argues that the rulings generally favor AI developers because they treat model training as analogous to reading a copyrighted work. Senior attorney Jason Henderson of JWL International notes that courts tend to reject training that directly competes with the original market, but are more permissive when the purpose is non-competitive. Both experts emphasize that the law has not kept pace with AI-driven data consumption.

TechCrunch

• Settlement amount: $1.5 billion (Anthropic).

TechCrunch

• Anthropic projects roughly $200 billion in annual revenue by 2028, according to Gellis.

TechCrunch

• U.S. copyright law has not been substantially revised since 1976, leaving judges to interpret outdated statutes for modern AI questions.

TechCrunch

The legal landscape remains unsettled. Courts continue to grapple with whether AI-generated works can be copyrighted, as highlighted by the Perlmutter decision that 100 % AI-generated content is not protectable. Determining the proportion of AI contribution in a work poses practical challenges, and many AI firms face pending lawsuits that could reshape the balance between authors’ rights and model training practices.

TechCrunch

• “I think it is generally good news for AI training that he looked at what was going on and really sort of thought it analogous to reading a copyrighted work as opposed to copying a copyrighted work,” — Cathy Gellis, an attorney with expertise in intellectual property, copyright, and technology

TechCrunch

• “Ross’s use is not transformative because it does not have a ‘further purpose or different character’ than Thomson Reuters’s,” — Judge Stephanos Bibas

TechCrunch

• “If you write your novel in [Microsoft] Word and run spell check, we kind of feel comfortable with the idea of saying that Word does not own your novel,” — Cathy Gellis, an attorney with expertise in intellectual property, copyright, and technology

TechCrunch

• “What you are seeing is that the initial opening volleys are being influential, and that influence itself could be undone if other courts decide different things, and it’ll take later states of litigation to figure out which one will prevail,” — Cathy Gellis, an attorney with expertise in intellectual property, copyright, and technology

TechCrunch