Anthropic Settles $1.5B Copyright Case Over Pirated Training Data
Settlement Covers Pirated Book Copies, Not All Training Data
AnthropIc agreed to pay $1.5 billion to resolve Bartz v. Anthropic, a class-action lawsuit brought by book authors in U.S. District Court for the Northern District of California The Leveraged Years. The settlement specifically covers piracy claims related to approximately 500,000 pirated titles that the company retained, not a blanket infringement judgment against all training data used to develop Claude or other AI models.
Judge William Alsup ruled in June 2025 that while training artificial intelligence on lawfully acquired books constituted fair use due to the transformative nature of the technology, retaining pirated copies of those works did not qualify for fair use protection The Leveraged Years. This distinction is critical: the court did not find that Anthropic illegally acquired all its training data, but rather that the company's possession of pirated datasets constituted copyright infringement.
Regulatory Requirement and Per-Work Payout
As part of the settlement, Anthropic must destroy the pirated datasets it downloaded, addressing the core legal violation. The $1.5 billion recovery translates to approximately $3,000 to $3,113 per work across the 500,000 titles, though this per-work figure remains an estimate pending final court approval Yahoo Finance.
The case reached preliminary approval in September 2025, but final approval by Judge William Martínez-Olguín remains pending as of July 2026 Clark Hill. The judge declined to approve five untimely opt-outs from the class, maintaining the settlement's scope.
Largest Copyright Recovery, Not Largest in U.S. History
The $1.5 billion settlement has been reported as one of the largest copyright recoveries on record, though sources stop short of confirming it as the single largest in U.S. history. The settlement marks a significant enforcement action against an AI company for data retention practices, even as courts have recognized that training on lawfully obtained material falls within fair use doctrine.
The outcome underscores the distinction between using legal training data—which courts have deemed a legitimate transformative use—and warehousing pirated copies, which remains infringing conduct. Anthropic's settlement obligation does not preclude the company from continuing to train AI models on books acquired through legal channels.
