Anthropic Settlement Approved by Judge
Anthropic settlement wins final approval and obliges $1.5 billion in scheduled payouts and data governance, tightening near-term cash flow for the company.
KEY TAKEAWAYS
- Judge approved a $1.5 billion settlement paid in four installments; $300M transferred, another $300M due within five days.
- Deal forces data-governance steps: destroy pirated files within 30 days and certify datasets used.
- Class covers roughly 482,000-500,000 registered works with about $3,000 per qualifying work.
HIGH POTENTIAL TRADES SENT DIRECTLY TO YOUR INBOX
Add your email to receive our free daily newsletter. No spam, unsubscribe anytime.
Anthropic P-ANTH won final court approval on July 20, 2026, for a class-action settlement resolving authors’ copyright claims over books used to train its Claude models. The Anthropic settlement requires scheduled payouts and data-governance steps that reshape the company's legal and operational exposure.
Final Approval, Payment Terms, and Compliance Obligations
U.S. District Judge Araceli Martinez-Olguin approved the settlement in Bartz v. Anthropic, filed in the Northern District of California’s San Francisco Division, rejecting objections that the recovery was inadequate. The agreement requires Anthropic to pay $1.5 billion in monetary consideration, structured in four installments. The company already transferred $300 million after preliminary approval and must pay another $300 million within five days of final approval, with the remaining balance phased over time.
The settlement obliges Anthropic to destroy the original pirated files within 30 days and certify which datasets were used in any commercially released models. It covers only past legacy uses of the works and does not grant a future license for continued use.
Class Composition and Payout Distribution
The certified class includes authors and book publishers whose registered U.S. works—identified by ISBN or ASIN—were alleged to have been downloaded and stored to train Claude. Eligible titles number roughly 482,000 to 500,000 registered works. Payments average about $3,000 per qualifying work and will be allocated according to ownership stakes among co-authors, publishers, and other rights holders. Plaintiffs’ counsel Lieff Cabraser said they intend to make distributions “as promptly as possible.” More than 91% of class members have claimed their payments.
The court awarded plaintiffs’ lawyers more than $101 million in fees, below requests that sought about $187.5 million.
Legal Rulings and Industry Implications
In June 2025, Judge William Alsup issued a summary judgment holding that training AI on lawfully acquired books can qualify as fair use, but using and storing pirated copies violates copyright law. Earlier findings alleged Anthropic downloaded millions of copyrighted books from piracy sites including Library Genesis and Pirate Library Mirror, retaining roughly 7 million titles in a central dataset used to train Claude. These claims formed the basis of the class allegations.
Because Anthropic settled, Alsup’s legal analysis remains a district-court precedent rather than binding appellate authority. The settlement resolves this case but leaves broader legal questions about AI training data unsettled. Some authors and rights holders opted out and continue separate actions.
The settlement’s financial schedule and its data-governance and certification requirements impose immediate compliance deadlines and operational constraints on Anthropic. Since the deal does not authorize future use of the covered books, the company must reassess data-sourcing and licensing practices for future model training.
