On July 20, 2026, Judge Araceli Martinez‑Olguin gave final approval to a deal that had been in the works for months. The money - $1.5 billion total - goes to a class of authors and book publishers who claimed Anthropic copied their work without permission to build its language models.
Earlier, former Judge William Alsup had given the preliminary green light before retiring. The settlement covers about half a million separate works, and each rights holder gets $3,000 per piece of writing used.
During the case, the court found that Anthropic built its training data by pulling books from pirate websites, which was illegal. The court determined that using copyrighted material to train AI models qualifies as fair use.
Get the market news that matters in a five-minute read with Market Briefs, our free daily newsletter
Anthropic opted to settle soon after the fair use decision, avoiding a jury trial that might have led to even higher damages.
Because Anthropic settled, no appeals court ever got to weigh in on the bigger question. That means the fair use ruling from this case does not set a binding precedent. Other judges in other cases can still decide things their own way.
And there are plenty of other cases. Google, Meta, Midjourney, and OpenAI all face similar lawsuits right now. Just last week, a brand‑new class action was filed against Google by a group that includes publishers Hachette, Cengage, and Elsevier, along with authors like Scott Turow and the group S.C.R.I.B.E.
The core issue in these lawsuits is whether it's fair use to train AI on copyrighted works. In the Anthropic case, the court said yes - but only for the actual training process, not for the method Anthropic used to obtain the books. The company was found to have downloaded works from pirate sites, which the judge called a clear infringement.
That distinction between lawful use and unlawful acquisition is likely to shape how other courts handle similar claims. For example, in the Google lawsuit filed by Hachette and Cengage, plaintiffs argue that mass scraping of copyrighted material - even from legitimate sources - still violates their rights. No binding precedent now exists, so outcomes could differ across jurisdictions.
The absence of a binding appellate decision means that similar lawsuits against other tech giants could produce different outcomes. Meanwhile, the $1.5 billion figure sets a potential floor for future settlements, as AI companies weigh the cost of litigation against the expense of licensing copyrighted works. This tension between fair use protection for training and the need to compensate creators will continue to drive legal strategy on both sides, with plaintiffs hoping to extract larger payments and defendants seeking to avoid sweeping penalties that could cripple AI development.
Join Market Briefs, our free daily newsletter, for a quick daily rundown of the markets
