Google Is Buying Korean Books for AI Training, Paper Says

Seoul Economic Daily reported on 18 October 2026 that “Google is also pursuing book purchase contracts with Korean publishers” (our translation), and that, according to the Korean publishing industry, “since the first half of this year, Google has been steadily negotiating with large Korean publishers to buy books for AI training”. Several industry officials told the paper: “we understand there are also cases where contracts have actually been completed”.

The paper said Google Korea did not reply to its questions, and that which publishers signed, and on what terms, has not been confirmed. Its sharper point is about price: the $3,000 per book Anthropic agreed to pay in the Bartz v. Anthropic settlement, which is damages for pirated copies, is being read in Korea as a price tag for AI training rights.

Business Pill · ROYALTIES

A short explainer on royalties: paying each time you use someone else’s work. It teaches the general idea only and says nothing about any company in this story.

What Seoul Economic Daily Reported

The report, by Park Yun-seon and Park Min-ju, says “an academic publisher is reported to be among those that signed” (our translation). It sets Google’s approach against Anthropic’s: it opens “amid controversy after AI company Anthropic was found to have bought a large number of books from a Korean book distributor”, and notes that Google chose publishers rather than distributors.

The paper writes that if publishers did not get each author’s consent, selling books as AI training data carries a high risk of legal dispute, and that Korea has no legal basis to regulate a distributor that sells content for AI training without the rights holder’s consent. It also recalls that “in July, Hachette Book Group and others filed a class action against Google in the US District Court for the Southern District of New York”.

What Was Posted on X

This is the post where we first saw the story. The summary is @choblin29’s own reading of the Seoul Economic Daily report, including the point that publishers fear Anthropic’s $3,000-per-book settlement could become the benchmark for AI licensing. The quotations in this piece come from our own translation of the Korean article, not from this post.

A Settlement Read as a Price Tag

The Bartz v. Anthropic settlement website says Anthropic agreed to a “non-reversionary Settlement Fund of one-billion-five-hundred-million dollars ($1,500,000,000)”, and that “It provides approximately $3,000 per work (not per Class Member)”. We covered the fight over how that pool is split.

Seoul Economic Daily writes that Anthropic’s “$3,000 (4.35 million won) per book in damages” is “not payment for AI training, yet it is being perceived as if it were a ‘price tag’ for publications used in AI training” (our translation). An official at an AI copyright company told the paper that “settlement money large foreign AI companies paid in lawsuits should not be confused with copyright fees”.

The same official warned: “if each negotiates separately without enough information or benchmarks, the price accepted by a publisher in urgent need could harden into the market standard”, adding that “there is a need to set standards for how licence prices vary with the scope of use” (our translation). The official also said Korea should prepare for “a market where AI references and cites books when it answers and settles payment repeatedly according to usage”.

The Other Price Points on Record

Citing the US magazine Vulture, Seoul Economic Daily writes that copyright holders of books in the Anthropic settlement get $3,000 per book, which in most cases is to be split 50/50 between author and publisher. On those terms the author’s half would be about $1,500 per book; that is our arithmetic.

Literary Hub reported one licence price in 2024. It wrote on 18 November 2024 that HarperCollins was “offering $2,500 (non-negotiable) for the right to use the book in an AI training deal” it had signed with an anonymous large tech company, according to an email its author Daniel Kibblesmith posted. No price for Google’s Korean contracts has been reported.

Negotiating Power

The AI copyright official told the paper that “many Korean publishers are relatively small, and the reality is that the size of any one publisher’s catalogue is not enough to negotiate on equal terms with a large AI company” (our translation).

A Korean conglomerate that builds AI models told Seoul Economic Daily that “Korean literary works cost more than general data when bought for AI training, so we are not training on them”, and that “for now we are training our AI mainly on data obtained from the government”. We reported earlier on a Korean lawmaker’s count of foreign models in government AI.

JoongAng Ilbo reported on 14 October 2026 that Kyobo Book Centre, Korea’s largest bookstore, had received “a bulk book order suspected of being for AI training” (our translation). It wrote that “since early this year, Kyobo Book Centre has received inquiries from companies based abroad, including in the US and UK, wanting to buy Korean-language books in bulk”, and that “about 10 companies have inquired so far”.

Asked about the end buyer, the companies told Kyobo that “information on the final customer is confidential and cannot be disclosed” (our translation), JoongAng Ilbo reported. One answered “it is not for AI training” while saying “the final customer is a data-related company”.

The Structural Read

As we read it, Google going to publishers rather than a distributor moves the deal to the party that holds contracts with authors, which is where Seoul Economic Daily says the consent question sits.

As we read it, the paper’s warning is about anchoring: with no disclosed licence price, the first terms a small publisher accepts may become the reference point, which is the risk the AI copyright official describes.

As we read it, the official’s suggestion of usage-based payments when AI cites books points to a second market after one-off training licences; neither Google nor any publisher has described such terms.

Three Implications

KOREAN PUBLISHERS The AI copyright official told Seoul Economic Daily that one publisher’s catalogue is not enough to negotiate on equal terms with a large AI company, and called for standards on licence prices.

AUTHORS Seoul Economic Daily writes that sales without each author’s consent carry a high risk of legal dispute, and that the Anthropic settlement is mostly split 50/50 between authors and publishers, citing Vulture.

KOREAN AI BUILDERS A Korean conglomerate told the paper it is not training on Korean literary works because they cost more than general data, and is using government data for now.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Four Intelligence Moats.

The framework lists four moats, one of them this: “The Corpus Moat — built by pretraining, eroding as the public web is exhausted and synthetic data closes gaps.”

As we read it, licensed books are one way to keep that corpus moat from eroding, which is why a price for them matters. The framework was published in July 2026 and does not discuss Google, Korean publishers or book licensing.

What Is Not Established

Google has not commented; Seoul Economic Daily says Google Korea did not reply. The report does not name the publishers, the number of contracts, the number of books or any price. The completed contracts rest on unnamed industry officials. We read the Korean report in full through its Daum syndication; quotations from it and from JoongAng Ilbo are our translation. The rules on data use differ by country; our piece on Meta’s UK terms covers one example.

The Bottom Line

Seoul Economic Daily reported on 18 October 2026 that Google has been negotiating with large Korean publishers to buy books for AI training since the first half of the year, with some contracts reportedly completed and an academic publisher reported to be among the signers. Terms are undisclosed. The paper reports that Anthropic’s $3,000-per-work settlement, which is damages, is being treated in Korea as a price tag.