Loading page…
The discovery gives affected writers a route to seek payment. It is not a new court finding that their books trained Claude.
Listen to this story
The Literature Translation Institute of Korea has identified 21 English-translated works by South Korean authors on the eligible-title list for Anthropic’s existing copyright settlement and notified their authors. The notices create a path to file claims, not an automatic payment: the roughly $3,000 per-work award may be divided among multiple copyright owners. Inclusion also does not establish that a book was used to train a commercially released Claude model, so rights holders must weigh the claim process against what the settlement actually covers.
The Authors Guild says the court granted final approval to the $1.5 billion Bartz v. Anthropic settlement on July 20, 2026.
The settlement covers specified past acquisition-and-copying claims through August 25, 2025; claims about AI outputs and future conduct remain available.
The institute named works by Han Kang, Shin Kyung-sook, Hwang Sok-yong and other Korean writers.
Korean writers including Han Kang and Hwang Sok-yong have been notified that their books qualify for compensation under Anthropic’s copyright settlement. On October 17, 2026, the Literature Translation Institute of Korea said it had identified 21 eligible works and contacted the authors, giving them a route to seek payment through the claims process.
The institute found the works on the compensation list published on the settlement website, according to ChosunBiz. The listed writers also include Kim Ae-ran, Kim Young-ha, Kim Sum, Shin Kyung-sook and Hwang Sun-mi. SBS described the affected books as English translations of works by South Korean authors.
An institute official told ChosunBiz that it would work with copyright-related institutions and support legal action if necessary to help authors obtain fair compensation. The notifications followed an October 16 appeal from the Korean Publishers Association, which called for transparent disclosure of AI companies’ use of copyrighted works and fair payment.
Those who actually made the books do not even know when, how, or which AI used their works.
Korean Publishers Association, quoted by ChosunBiz
The discovery concerns an existing settlement, not a newly negotiated deal. According to the Authors Guild, a federal court granted final approval of the $1.5 billion settlement in Bartz v. Anthropic on July 20, 2026. ChosunBiz puts the number of compensation-eligible titles at about 500,000.
Finding rights holders has itself required a large outreach effort. In its July account, the Guild said settlement notices went by mail and email to nearly 595,000 potential class members. They reached rights holders associated with 99.5 percent of listed works. The administrator assembled contact details with help from writers’ organizations, the Authors Registry and more than 170 publishers.
Korean authors must participate in the claims process to receive compensation. The approximately $3,000 per-work award is not necessarily a payment entirely to one author. The court-approved allocation plan divides a valid claim’s per-work share among copyright owners, using eligible default splits or percentages established by publishing contracts, the Guild explains.
Disagreements over who owns a share of a book’s award can go to a court-appointed Special Master, Theodore K. Cheng. Publishing agreements and other submissions in those disputes remain confidential and under seal. Identifying an eligible title and establishing how its payment should be divided are therefore separate steps.
Legal fees also affect how much remains for rights holders. The Guild says the court cut class counsel’s fee award to about $101.56 million, or 6.8 percent of the fund, below their reduced request for 12.5 percent. Ten percent of that award is withheld pending a post-distribution accounting. Counsel also committed to receiving fees in step with payments to class members, rather than simply when money enters the fund.
Money saved on fees stays in the fund for authors. If funds remain after valid claims are paid, the parties anticipate another distribution to class members. A final balance too small to distribute economically would go to charitable purposes, subject to court approval.
The settlement’s legal scope is narrower than a finding that every listed book trained Claude. It resolves claims about Anthropic’s past acquisition and copying of the covered works through August 25, 2025, according to the Authors Guild. Inclusion on that list does not itself establish use in a commercially released model.
The Guild says Anthropic represented that neither the Library Genesis nor Pirate Library Mirror dataset, nor any portion of either, was in the training corpus of its commercially released large language models. The settlement nevertheless requires destruction of original files downloaded from those sources and copies derived from them, subject to legal preservation obligations.
Nor does accepting the settlement release every possible copyright claim. Claims concerning AI outputs and future conduct remain available, the Guild says. Works absent from the settlement list also retain their claims, even if they otherwise appear to meet the class criteria. The list determines which past acquisition-and-copying claims are covered; it does not settle the rights attached to every book an author has written.
Loading discussion...
Join the conversation
Explain who should bear the work of establishing payment rights.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.