A San Francisco judge just made AI copyright fights a $1.5 billion problem

A ruling with industry-wide stakes

A San Francisco federal judge approved Anthropic’s $1.5 billion settlement, the largest known U.S. copyright settlement to date. It creates a major new payment benchmark for creators.

The ruling warns AI companies that training data has a legal supply chain. Obtaining material unlawfully can bring enormous costs, even when model training qualifies as fair use.

The case began with three authors

Authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson sued Anthropic in August 2024. They said protected books were copied without permission while the company built systems behind Claude.

The lawsuit later became a class action for qualifying copyright owners. Their books appeared in datasets Anthropic obtained from LibGen and PiLiMi, two online shadow libraries.

Millions of books entered one library

Court records said Anthropic downloaded more than 7 million pirated books and stored them in a central digital library. The company also bought printed books and scanned every page.

That mixed sourcing became crucial. Judge William Alsup treated legally acquired copies differently from files taken through LibGen and PiLiMi, even when both collections supported development.

Training and piracy were separated

Judge William Alsup ruled in June 2025 that training AI models on legally acquired books could qualify as fair use. He refused to excuse Anthropic’s permanent library of pirated copies.

That legal split remains central today. The agreement does not declare AI training illegal; it resolves claims tied mainly to unauthorized downloading, copying, and storage.

Little-known fact: In June 2025, AI-related positions represented 42% of all tech job postings in the San Francisco Bay Area, up from 20% in mid-2022. The region recorded 11,400 AI-related openings that month, its highest level on record.

The deal took nearly a year

The parties reached a settlement in August 2025, before a damages trial scheduled for December. Judge Alsup granted preliminary approval that September.

Review continued through objections, claims processing, and a May 2026 fairness hearing. On July 20, Judge Araceli Martinez-Olguin approved the agreement and entered judgment in San Francisco federal court, ending the class case.

Nearly half a million works qualify

The final works list contains 482,460 eligible books, far fewer than the millions of downloaded files. A title generally needed an ISBN or ASIN and timely U.S. registration.

Only works listed by the settlement administrator are covered. Books outside that database retain separate claims, even when authors or publishers own other titles in the agreement.

Little-known fact: San Francisco-based AI companies received $103 billion in venture capital from 2020 through early 2025. They accounted for 43% of all U.S. AI venture funding during that period.

The headline payout needs context

The settlement estimates roughly $3,000 per eligible work, before legal fees, administrative costs, expenses, and interest adjustments. Payment attaches to the title, not automatically to every person connected with it.

Authors, publishers, coauthors, or rightsholders may divide one award. For many trade and university press books, the default allocation is a 50-50 split between author and publisher.

Most eligible works were claimed

By April 16, valid claims covered 440,490 of the 482,460 works on the official list, a 91.3% participation rate. That broad response strongly supported final approval.

The judge described class support as overwhelmingly favorable. Only 350 timely opt-outs, covering 1,802 works, were recorded, while 54 objections or comments reached the court before the final ruling.

The judge cut the lawyers’ request

Class lawyers first sought $300 million, then lowered their request to $187.5 million after judicial scrutiny. Judge Martinez-Olguin approved $101.56 million, about 6.8% of the fund.

She also approved expenses and a future-cost reserve. The court will withhold 10% of fees until lawyers submit a post-distribution accounting, protecting the class against an excessive eventual payment.

Anthropic must delete pirate files

Money is only one part of the agreement. Anthropic must destroy files downloaded or torrented from LibGen and PiLiMi, plus copies originating from them.

The company must certify the deletion after final judgment. Scanned copies made from purchased books are excluded, reflecting the court’s distinction between lawful acquisition and piracy while preserving digitized purchased-book scans.

Trial risk could have been enormous

Without settlement, a jury could have considered statutory damages across hundreds of thousands of works. Federal copyright law generally allows $750 to $30,000 per infringed work.

Awards can reach $150,000 for willful infringement. Reuters reported Anthropic’s exposure could have reached hundreds of billions of dollars, so the $1.5 billion fund replaced a larger, uncertain risk.

Some authors are still suing

The settlement does not end every copyright claim against Anthropic. Rightsholders who opted out may still pursue separate cases, while excluded works are not released.

More than 100 authors filed another lawsuit in June 2026, this time in Northern California, involving more than 500 books. They seek individual statutory damages instead of accepting the class settlement’s payment structure.

Could Gen Z’s move away from San Francisco reshape America’s next major tech hubs? Take a closer look at how younger workers, startup talent, and investment are helping drive new technology growth in other states.

Other AI companies now face a benchmark

The agreement is not a binding verdict against OpenAI, Google, Meta, or rivals. Still, it gives courts, creators, and investors an example of a training-data dispute’s price.

Future cases will turn on facts. Dataset origin, copyright registration, market harm, licensing history, and recordkeeping could decide whether a company pays nothing, negotiates licenses, or faces damages.

Could this $1.545 million San Francisco apartment reveal how modest OpenAI’s earliest beginnings really were? Take a closer look at the city home linked to the company’s first work, its startup history, and the surprising price attached to a piece of San Francisco AI history.

Should AI companies face billion-dollar penalties for using copyrighted books without permission, or should training data fall under fair use? Share your thoughts in the comments below.

This slideshow was made with AI assistance and human editing.

Read More From This Brand:

• Back-to-back power disruptions have frustrated residents and businesses in San Francisco
• Even six-figure earners face unique financial trade-offs in San Francisco
• A sudden school shutdown leaves San Francisco families scrambling as a $395 million church settlement looms