Quick Take
- A US federal judge approved Anthropic’s $1.5 Bn (Rs 14,445 Cr) copyright settlement on July 20, 2026.
- Roughly 500,000 pirated books qualify, with authors and publishers receiving about $3,000 (Rs 2.89 Lakh) per work.
- The court cut lawyer fees to $101.5 Mn and set no binding precedent for other AI cases.
In This Article
A US federal judge granted final approval to the Anthropic copyright settlement worth $1.5 Bn (Rs 14,445 Cr) on July 20, 2026, ending a class action brought by authors over pirated books used to train the Claude chatbot.
US District Judge Araceli Martínez-Olguín signed the order in San Francisco. The deal covers roughly 500,000 works and pays about $3,000 (Rs 2.89 Lakh) for each infringed book. Lawyers on both sides call it the largest copyright recovery in United States history. Anthropic must also destroy the pirated files it downloaded from shadow libraries.
StartupFeed Insight
The number that matters is not $1.5 Bn, it is $3,000 per book. That figure now sits in every AI licensing negotiation as an anchor price, four times the $750 statutory minimum a court would normally award. Anthropic settled rather than appeal, so the fair use ruling never became binding law. Indian AI founders training on scraped corpora should read this as a pricing signal, not a safety signal. StartupFeed expects at least three Indian publishers or news groups to send formal data-licensing demands to domestic LLM builders before March 2027. By Avinash.
Anthropic Copyright Settlement: The Numbers
The Anthropic copyright settlement is a $1.5 Bn class action fund created to compensate authors and publishers whose books Anthropic downloaded from pirate libraries. Judge Martínez-Olguín inherited the case after Judge William Alsup retired at the end of 2025, court records show.
| Metric | Detail | Notes |
|---|---|---|
| Total settlement fund | $1.5 Bn (Rs 14,445 Cr) | Largest known US copyright recovery, per class counsel |
| Payout per work | About $3,000 (Rs 2.89 Lakh) | Four times the $750 ordinary statutory minimum, court order |
| Works covered | Roughly 500,000 | Shared between authors and rights-holding publishers |
| Attorney fees awarded | $101.5 Mn (Rs 977 Cr) | Cut from the $187.5 Mn requested, court order |
| Named plaintiff award | $15,000 each | Reduced from $50,000 sought, per Authors Alliance |
| Final approval date | July 20, 2026 | Judge Araceli Martínez-Olguín, Northern District of California |
The fee cut is the sharpest detail. Class counsel first asked for $300 Mn, trimmed it to $187.5 Mn, and received $101.5 Mn. That works out to roughly $2,953 per hour across 34,381.6 credited attorney hours, according to Authors Alliance analysis of the final order.
About Anthropic
Anthropic is a US artificial intelligence safety company founded in 2021 by Dario Amodei, Daniela Amodei and five other former OpenAI researchers. Headquartered in San Francisco, it builds the Claude family of large language models and sells access through an API and enterprise subscriptions. Backers include Amazon, Google and Spark Capital. Claude competes directly with OpenAI’s ChatGPT and Google’s Gemini.
What does this mean for AI companies?
The Anthropic copyright settlement resolves how the books were obtained, not whether AI training is legal. Judge Alsup ruled in June 2025 that training Claude on lawfully bought books counted as fair use, but that storing over seven million pirated copies in a permanent library did not. Anthropic admitted downloading from Library Genesis and Pirate Library Mirror.
We reached this settlement in 2025, after the court’s landmark ruling that training AI on books is fair use under copyright law, which remains the law today, said Aparna Sridhar, deputy general counsel, Anthropic.
Because Anthropic settled, the fair use finding never reached an appeals court. It stays a single district court decision with no binding force. Judges hearing cases against Google, Meta, OpenAI and Midjourney remain free to rule differently on their own facts. The settlement also does not license Anthropic’s future training runs.
Why should Indian AI founders care in 2026?
India has no settled rule on whether AI training on copyrighted text is lawful. The Copyright Act, 1957 has no text and data mining exception of the kind the European Union and Japan provide. A commerce ministry panel and the Department for Promotion of Industry and Internal Trade have both examined the question, and the DPIIT policy framework remains the reference point for any change.
Indian LLM builders working under the IndiaAI Mission face a practical problem. Most open Indic-language corpora are assembled by scraping, and provenance records are thin. The IndiaAI Mission dataset platform is trying to fix that with vetted public datasets. Founders who cannot show a clean acquisition trail now carry a costed risk, roughly Rs 2.89 Lakh per work if an Indian court ever borrows the US anchor.
How do rival AI copyright cases compare?
| Company | Case status | Exposure |
|---|---|---|
| Anthropic | Settled, approved July 20, 2026 | $1.5 Bn (Rs 14,445 Cr) paid |
| OpenAI | Multiple suits ongoing | Undetermined, no settlement |
| Meta | Author and publisher suits ongoing | Undetermined, contested |
Anthropic also faces a separate music copyright suit filed in January 2026 by Universal Music Group, Concord and ABKCO over song lyrics used in training. What separates Anthropic from its rivals is simple: it is the only major AI lab that has now written a cheque and set a public per-work price.
What’s Next
Payments start one business day after the effective date, which arrives once the 30-day appeal window closes. If neither side appeals, money should reach class members by late August 2026. Authors who opted out are litigating separately, and those suits stay live. The bigger question sits with Indian regulators: will DPIIT propose a text and data mining exception before the next Copyright Act review? What would a fair per-work rate look like in rupees?
Frequently Asked Questions
Written by Avinash. Have a tip? Write to us at editorial@startupfeed.in.
