Anthropic’s proposed $1.5bn copyright settlement could bring a major legal dispute with authors closer to an end, but its final approval status has not been independently confirmed. A firm timeline for distributing payments also remains unclear.
The case is significant because it separates two questions that are often bundled together in arguments about generative AI: whether copyrighted books can be used to train a model and whether the company lawfully obtained those books.
Authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson filed the lawsuit in 2024, alleging that Anthropic used pirated copies of copyrighted works while developing Claude. Their complaint focused not only on AI training but also on the way the underlying material was acquired.
AI training and book acquisition are different issues
Anthropic maintains that training AI models on books is protected by fair use. A court ruling supporting that position has been reported in connection with the case, although the details and status of that decision have not been independently verified.
The separate acquisition issue is potentially more damaging. The proposed class is said to include copyright owners whose books appeared in the shadow libraries LibGen and PiLiMi and were downloaded by Anthropic. Those allegations have not been established here as confirmed facts, but they form the central dividing line in the dispute.
That distinction matters well beyond Anthropic. A company could potentially have a defensible argument for transforming lawfully obtained material during model training while still facing liability for acquiring unauthorized copies. The settlement therefore should not be read as a simple ruling that all book-based AI training is either legal or illegal.
How the proposed payments compare
The proposed fund totals $1.5bn. If divided among approximately 500,000 eligible works, the headline calculation comes to roughly $3,000 per work. Both the number of covered works and the resulting average remain subject to the settlement process, however, and should not be treated as a guaranteed payment for every claimant.
The reported per-work figure has also been compared with statutory damages for wilful infringement, but that comparison has not been independently confirmed. Eligibility rules, duplicate claims and administrative costs could all affect what individual copyright owners ultimately receive.
Anthropic said more than 91% of the authors and publishers covered by the proposed settlement had claimed a share of the payment. The company also said it was looking forward to bringing the matter to a close.
What remains unresolved
Three details will determine the settlement’s practical impact:
- Final legal status: Judicial approval has been reported, but that status remains unconfirmed.
- Eligibility: Copyright owners will need to know which works qualify and how competing claims are handled.
- Payment timing: No firm public distribution schedule has been confirmed.
The broader takeaway is less about the size of the proposed deal than the legal boundary it exposes. The debate over whether AI training qualifies as fair use does not automatically resolve how training material may be collected. For AI developers, authors and publishers, those remain distinct questions with very different consequences.