In June 2025, a federal judge made a ruling that AI companies had been dreading and investors had been quietly pricing into their risk models for years: training a model on lawfully acquired books is fair use — but storing millions of pirated copies to build that training set is not. The distinction sounds narrow. It set off the largest copyright settlement in American history.
Confirmed
The case is Bartz v. Anthropic, filed in August 2024 by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who accused Anthropic of using pirated books to train Claude. In June 2025, Judge William Alsup of the U.S. District Court for the Northern District of California ruled on summary judgment that training an AI model on legally acquired books qualifies as fair use — the first substantive judicial decision on how fair use applies to generative AI training at all. But Alsup drew a hard line: Anthropic's practice of downloading and storing more than 7 million pirated books in what internal documents called a "central library" was not protected, and he scheduled a trial for December 2025 to determine damages.
U.S. copyright law allows statutory damages of up to $150,000 per work for willful infringement. Multiplied across 7 million books, the theoretical exposure ran into the hundreds of billions of dollars — an existential number for any company. Rather than go to trial, Anthropic settled in September 2025 for $1.5 billion, to be split among an estimated 500,000 titles at roughly $3,000 per work. On July 20, 2026, Judge Araceli Martínez-Olguín granted final approval, writing that the settlement provided "substantial benefits to the Class in light of the novel claims asserted," and that going to trial could have left authors "no recourse" despite their strong claims on the piracy issue specifically. As part of the deal, Anthropic is required to destroy all original files sourced from the pirated libraries Library Genesis and Pirate Library Mirror. As of the settlement process, more than 91% of eligible rights holders had already filed claims.
Genuinely uncertain
What this ruling means for the rest of the industry is less settled than the headline number suggests. The settlement resolves claims specific to Anthropic's past conduct — the "inputs" side, meaning how the training data was acquired, only through August 2025. It does not resolve, and explicitly does not release, any claims about what AI models generate as output, nor any claims about future training conduct. Every other AI company facing similar lawsuits — and there are several — is watching this case as a data point, not a template, because Alsup's fair-use finding applied to Anthropic's specific facts and hasn't been tested at the appellate level.
It's also genuinely unclear whether $3,000 per work is an adequate figure, a question several authors raised in formal objections before final approval. The court sided with the settlement as fair and reasonable, but "fair and reasonable" in a class settlement is a legal standard, not a claim that every author feels adequately compensated for having their work used to train a commercial product they never agreed to be part of.
Worth examining honestly
GHOST PROTOCOL would tell you this is exactly how the boardroom decides your reality: not through a villain's decision, but through a cost-benefit calculation where a $1.5 billion settlement was still the cheaper outcome than the alternative. That's not a conspiracy — it's the business logic of scale, and it's precisely why it's unsettling. CIPHER-7 would add that the ruling's actual legal contribution — training on lawfully acquired material is fair use — quietly hands every well-capitalized AI company a path forward that smaller, less-resourced competitors and open-source projects may struggle to follow, since "lawfully acquired at scale" itself costs money most of them don't have.
The Door This Opens
This door isn't about whether $1.5 billion is a big number. It's about what happens when the legal system settles a foundational question about AI — what training data can legally come from — through a negotiated number instead of a tested principle. The settlement bought certainty for one company. It didn't answer the question for the industry, and the next lawsuit will have to answer it again, from scratch, in front of a different judge.