Judge Rejects Proposed Anthropic Settlement Over AI Piracy Claims

Ashley Belanger of Ars Technica reports that a federal judge has rejected the proposed $1.5 billion Anthropic settlement to resolve claims that the company illegally used millions of books to train its AI systems.

US District Judge William Alsup criticized the agreement as incomplete and potentially unfair to authors, warning that it could release the company from massive copyright liability without resolving key procedural issues.

Alsup raised concerns that the settlement might allow Anthropic to escape accountability at a low cost while providing inadequate compensation to the affected authors.

The case stems from a certified class action of up to seven million claimants whose works were allegedly downloaded without authorization.

The settlement, however, would cover fewer than 500,000 works. The Authors Guild attributes this to duplicates, foreign editions, unregistered works, and books that do not meet class criteria.

Although the Copyright Act allows statutory damages of up to $150,000 per willfully infringed work, authors reportedly accepted lower payouts to avoid the years of litigation and appeals that would have ensued.

The Guild emphasized that the deal was intended to provide faster, guaranteed payments while sending a message to AI companies about the cost of using copyrighted works without licenses.

In denying preliminary approval, Alsup cited the absence of essential elements, including finalized lists of works and class members, a defined claims process, and rules for resolving ownership disputes.

He warned that incomplete notice procedures could trigger future claims despite Anthropic’s payout, leaving courts mired in disputes for years.

Alsup has ordered that the outstanding details be finalized by late September, setting a potential timeline for reconsidering settlement approval in October.

Lawyers should monitor how this intervention alters the settlement’s structure. His insistence on rigorous claims administration, clear opt-in rules, and comprehensive release terms may set procedural expectations in future copyright class actions involving AI training data.