Is training AI on copyrighted books legal

The gist

• Judge William Alsup ordered Anthropic to pay a $1.5 billion copyright settlement to writers, but ruled the AI training itself was lawful and penalized the company for pirating books from illegal shadow libraries.
• Alsup wrote that "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."
• Attorney Cathy Gellis says the ruling favors AI companies, noting Anthropic is projecting roughly $200 billion in annual revenue by 2028, and that "copyright law hinges on copying, but it doesn't hinge on using the work."
• US copyright law has not been updated since 1976, leaving judges to apply 50-year-old guidelines to AI cases.
• In Thomson Reuters v. Ross Intelligence, Judge Stephanos Bibas found no fair use, writing that "Ross's use is not transformative because it does not have a 'further purpose or different character' than Thomson Reuters's."
• In Thaler v. Perlmutter, the court ruled that a 100% AI-generated work is not copyrightable, raising questions about how to prove what share of a work was AI-assisted.

Discussion