A Napster Moment for AI? (cepa.org)

Anthropic agreed to pay at least $1.5 billion to settle a copyright suit brought by authors after the company used copyrighted books from “shadow libraries” to train its Claude models. The deal—if approved—would be the first billion-dollar payout by an AI firm, and would require mass destruction of tainted training data under court supervision and a framework for compensating rights holders. Judge William Alsup has blocked preliminary approval, demanding an ironclad list of allegedly pirated works and a transparent claims process; without that, the case heads to a December trial that could expose Anthropic to larger statutory damages. The settlement highlights per-work liability math that matters: Anthropic’s offer roughly equates to $3,000 per work (far above the $750 statutory baseline and the $200 innocent-infringement floor) and flags the real risk of costly litigation or court-ordered data deletion. For AI/ML practitioners and policymakers, this is a watershed moment: courts are willing to force meaningful payouts and dataset remedies, which pushes the industry toward robust copyright-compliance, provenance tracking, and negotiated licensing markets. But text licensing is technically and institutionally harder than music—works are fragmented across publishers, independents, and estates—so building registries, standardized metadata, and legal clarity will be necessary. With Congress unlikely to act quickly, expect precedent-driven settlements, tighter data governance, and accelerated investment in auditable training pipelines as firms seek to avoid Anthropic’s fate.