Bartz v. Anthropic — Court Approves $1.5 Billion AI Copyright Settlement and Enters Judgment

Background

Authors and copyright owners sued Anthropic over copies of books obtained from the LibGen and PiLiMi shadow libraries and used in developing its artificial-intelligence systems. After earlier rulings separated lawful training uses from claims based on acquiring pirated copies, the parties negotiated a non-reversionary $1.5 billion class settlement covering 482,460 works. The court preliminarily approved the deal, directed a broad notice program, and later considered the final-approval record, objections, opt-outs, fees, expenses, and service awards.

The Court’s Holding

The court granted final approval and entered judgment, finding the settlement fair, reasonable, and adequate under Federal Rule of Civil Procedure 23. Notice reached potential rightsholders associated with virtually all listed works, 91.3 percent of the works had been claimed by mid-April, and only 350 timely opt-outs covering 1,802 works were submitted. The court found no signs of fraud or collusion and concluded that the approximately $3,000 estimated payment per work provided substantial value in light of the cost and uncertainty of continued litigation.

The approved release covers claims tied to Anthropic’s past acquisition and copying of works on the settlement list, but it does not release claims concerning AI outputs or conduct occurring on or after August 25, 2025. The court overruled the objections, approved the allocation and claims-dispute process, and granted the fee request only in part, reducing the requested award while approving expenses and representative service awards.

Key Takeaways

• The $1.5 billion fund is non-reversionary, so money does not return to Anthropic if initial distributions leave a balance.
• The settlement is limited to listed works sourced from LibGen or PiLiMi and does not resolve future-conduct or AI-output claims.
• The court treated the high claims rate, very small opt-out rate, extensive discovery, and litigation risk as strong evidence that the deal delivered meaningful class relief.
• Final approval converts the negotiated settlement into an enforceable judgment and ends the covered action once the agreement becomes effective.

Why It Matters

This judgment closes the largest known U.S. copyright class settlement involving generative-AI training materials and establishes a concrete monetary benchmark for resolving claims based on pirated training copies. Its carefully limited release also preserves a dividing line between liability for acquiring unauthorized source copies and still-unresolved questions about model outputs and later AI practices.