← All news·2026-08-23·3 min read

Training AI on Copyrighted Books: Two Courts, Two Contradictory Rulings

Judge Alsup ruled that Anthropic's AI training on books is legal but ordered $1.5B in damages for piracy. A separate court reached the opposite conclusion in the Thomson Reuters case — and both verdicts are now active precedents.

aicopyrightregulationlaw

In 2025, Judge William Alsup decided in the Anthropic case that it’s OK to train an AI model on copyrighted material (books). It’s like “a writer reading other writers” not “copying”. But Anthropic had to pay $1.5B for using pirated copies of those books, which is illegal even if training an AI model is not.

In the same year, another judge said the exact opposite in Thomson Reuters v. Ross Intelligence. Judge Stephanos Bibas held that “using copyrighted material to create a rival product is not fair use.” Why? Because this doesn’t “transform” the original material, it just replaces it. So we have two conflicting precedents.

This is not about training an AI model, but about what happens after training: whether the AI model harms the market of the original creator. Copyright law was created 50 years ago, when there were only books and vinyl records. We don’t have a clear, consistent precedent yet that applies to all cases involving AI models.

Source: techcrunch.com

Free course

Stop reading about AI — start building with it

The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.

Start free →
EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Founder & CEO, Dad4Kids