Judge Pauses Final Approval of Anthropic's $1.5 Billion Copyright Settlement After Author Objections

A federal judge delayed final approval of Anthropic’s $1.5 billion settlement over alleged use of books in AI training after authors objected to distribution terms and a roughly $320 million fee request;

A federal judge has paused final approval of a $1.5 billion settlement resolving claims that Anthropic used hundreds of thousands of books to train its AI models, citing objections from authors who say the deal shortchanges creators. U.S. District Judge Araceli Martinez‑Olguin asked parties to address those objections rather than immediately approve the agreement, signaling the court will scrutinize both the settlement’s distribution plan and the size of attorneys’ fees.

The settlement covers more than 480,000 works alleged to have been used in training. Plaintiffs’ counsel has requested roughly $320 million in legal fees from the fund, while objectors say individual awards would average only about $3,000 per claimant under the current structure. Objecting class members argue that, as drafted, the payout mechanics favor counsel and leave many entitled authors undercompensated.

Authors’ representatives reported to the court that claims had been filed for over 92% of the works included in the settlement, but objectors counter that many eligible authors have not registered claims and may never be compensated. One objector, Pierce Story, contends the fee request effectively compensates counsel at an estimated $10,000–$12,000 per hour and invoked an 8th Circuit observation from a T‑Mobile case to argue against what he describes as unreasonable fee awards in class actions.

Objectors have proposed reducing attorneys’ fees to increase individual payments. Story offered an illustrative reallocation showing that lowering counsel’s payout to $70 million — still characterized as generous in his example — would boost individual plaintiff payments by nearly 25%. The judge’s decision to probe these arguments opens the possibility that fee terms and distribution mechanics could be renegotiated before any final sign‑off.

Several class members warned that approving terms perceived as unfair could leave the settlement vulnerable to appeal. Meanwhile, a group of 25 class members who opted out has filed a separate lawsuit, underscoring that litigation over unauthorized use of copyrighted works for AI training is still unfolding even as parties pursue a historic resolution. For developers, publishers and legal teams, the pause underscores continuing uncertainty about remedies for large‑scale use of copyrighted texts in model training. The court’s scrutiny of attorney fees and claimant registration may set a precedent for how future AI‑related copyright settlements are evaluated, keeping compliance and risk management priorities on the agenda for both builders and content owners.