The Anthropic $1.5B Ruling Didn't Hurt AI Training Rights , It Clarified Them

1. The Anthropic $1.5B Ruling Didn't Hurt AI Training Rights , It Clarified Them

Last year, Judge William Alsup ordered Anthropic to pay a $1.5 billion copyright settlement to a group of authors whose works were used to train the company's Claude models. The penalty was not for training on copyrighted material. Alsup ruled that AI training itself was lawful. What he penalized was Anthropic's sourcing method: pulling books from illegal shadow libraries. In his ruling, Alsup compared an LLM ingesting text to a writer studying literature, writing that Anthropic's models trained on those works "not to race ahead and replicate or supplant them , but to turn a hard corner and create something different."

The strategic read here favors AI labs, not authors. A $1.5 billion fine against a company projecting roughly $200 billion in annual revenue by 2028 is a rounding error, not a deterrent. What the ruling actually produced is a cleaner legal frame: training on copyrighted text may survive fair use analysis, provided the sourcing is clean and the purpose is transformative rather than directly competitive. IP attorney Cathy Gellis told TechCrunch the ruling is "generally good news for AI training," because Alsup treated ingestion as analogous to reading, not copying, and copyright law turns on copying. Labs with licensed or legitimately sourced data now have a stronger position than they did before this ruling existed.

The unresolved edge is competitive purpose. Attorney Jason Henderson notes courts are "all over the place" in their reasoning, but a pattern is forming: training that directly competes with the source material draws judicial skepticism, while training aimed at building something different does not. Copyright law has not been updated since 1976. Until Congress acts or a higher court draws a firmer line, the question of what counts as "transformative" will keep getting answered case by case, and each ruling will reset the risk calculus for every lab building on scraped data.

Source: Is it legal to train AI models on copyrighted books? It's complicated