Anthropic's $1.5B fine was piracy punishment, not an AI training ban

The $1.5 billion copyright settlement Anthropic agreed to pay last year looked like a massive win for authors. Judge William Alsup ordered the payment to a group of writers whose books were used to train Claude. But the ruling wasn’t what it appeared to be.

Alsup actually decided that training AI models on copyrighted works is lawful. The fine targeted something else entirely: Anthropic’s use of pirated books from illegal shadow libraries. The judge compared how an LLM ingests text to how a writer studies literature, writing that Anthropic’s models trained on works “not to race ahead and replicate or supplant them — but to turn a hard corner and create something different.”

For AI companies, that distinction matters enormously. Attorney Cathy Gellis pointed out that $1.5 billion is pocket change for a company projecting roughly $200 billion in annual revenue by 2028. The real victory was the legal principle. Copyright law hinges on copying, not on reading or consuming a work — and Alsup essentially said training is closer to reading.

The problem is that copyright law hasn’t been updated since 1976. Judges are now applying half-century-old guidelines to questions that didn’t exist when the law was written. Fair use arguments hinge on whether a use is “transformative,” and courts have been inconsistent. In Thomson Reuters v. Ross Intelligence, Judge Stephanos Bibas ruled that training on legal content to build a competing platform was not fair use. The deciding factor? Whether the AI product directly competes with the original work.

That competitive question remains unresolved for authors who argue chatbots compete with them by generating synthetic books. Meanwhile, a separate case — Thaler v. Perlmutter — established that purely AI-generated work can’t be copyrighted at all, raising thorny questions about how to prove what percentage of a work involved AI assistance. The legal landscape is fragmented, and the next few rulings will likely determine whether AI training on copyrighted material remains viable or becomes prohibitively expensive.