Google Is Negotiating to Buy Korean Books for AI Training, Seoul Paper Reports

Google has been in talks with large South Korean publishers to buy books for training artificial intelligence, according to a report by Seoul Economic Daily, which cites unnamed sources in the publishing industry. Neither Google nor any publisher has confirmed a deal. No prices have been reported.

The report was published on 18 October at 18:07 Korea time (09:07 UTC, 10:07 UK time). Its English edition says it is “AI-translated from Korean”, so the quotes below are from that translation. TSN also checked the Korean original, which says the same thing.

What the report says

“Google has been negotiating with major Korean publishers to buy books for AI training since the first half of this year, according to publishing industry sources on the 18th,” Seoul Economic Daily reports.

“Several industry officials said they understood that some contracts had already been completed,” it adds. The paper says it could not confirm which publishers had signed or on what terms, “though academic publishers were reportedly among them.”

“Google Korea did not respond to a request for comment,” according to the report.

So this is an account from unnamed sources. It is not a confirmed deal. No publisher has been named, no price has been given, and Google has not commented.

The Anthropic background

According to the report, Google’s approach follows “the controversy over AI company Anthropic’s bulk acquisition of Korean books from a local book distributor.” Seoul Economic Daily says that after AI companies were caught training models on illegally downloaded books, they “shifted toward signing direct purchase contracts.”

The paper says Korean law sets no rules for distributors that sell content for AI training without the copyright holder’s consent. Google is going to publishers rather than distributors, the report says, “but the underlying issue is the same”: if publishers have not got individual consent from authors, selling books as training data “carries a high risk of legal disputes.”

The paper also notes that Google faces its own copyright case over AI training in the US. According to the report, Hachette Book Group and others filed a class action against Google in the US District Court for the Southern District of New York in July.

The $3,000 worry

The main concern in the report is price. It says Anthropic’s “$3,000-per-book settlement over pirated Korean literary works — which was damages for piracy, not payment for AI training — is being treated as a de facto price tag for AI training material”. The Korean original puts the figure at 4.35 million won per title.

The figure comes from Bartz v. Anthropic, a US class action over pirated books. In its order granting final approval on 20 July 2026, the federal court in Northern California described a “$1.5 billion” settlement fund. It said “the estimated per-work payment of approximately $3,000” and, quoting an earlier ruling, called it “a per-work award of about $3,000, less costs and fees.” The settlement website says the first payments are expected on or before 15 November 2026.

Citing the US magazine Vulture, Seoul Economic Daily says most rights holders in that settlement are splitting the money 50-50 between author and publisher. It adds that some publishers demanded all of it, and some authors received nothing because their copyrights were registered too late.

An unnamed official at a company working on AI copyright issues told the paper that the settlements large AI companies have paid in lawsuits should not be confused with copyright fees. “If each party negotiates without sufficient information or standards, a price accepted by a publisher in urgent need could harden into the market standard,” the official said. According to the report, the same official also pointed out that many Korean publishers are small, which makes it hard for them to negotiate as equals with large AI companies.

Korean AI firms and pending bills

The report says Korean AI developers face a different problem. An official at a Korean conglomerate building AI models said the company is not training on Korean literary works because that data costs more than general data. “For now, we are training our AI mainly on data obtained from the government,” the official said.

A publisher quoted in the report said measures prepared by the industry would mean nothing unless legislation passes. “Even if everything is taken from Korea by companies abroad, there is nothing we can do,” the publisher said.

The paper says several amendments to Korea’s Framework Act on Artificial Intelligence Development and Trust have been introduced in the National Assembly. They “would require disclosure of the sources of generative AI training data and set out procedures for copyright holders to verify such data”, but they “remain pending”. These are proposals, not law.

What we don’t know

• Whether Google has actually signed any contracts. That claim comes from unnamed officials, and Google Korea did not respond.
• Which publishers are involved, beyond the report that academic publishers are among them.
• Any prices or terms.

What this means

• Seoul Economic Daily reports that Google has been negotiating since the first half of 2026 to buy books from major Korean publishers for AI training, and that some contracts may already be done.
• Korean publishers are worried that the roughly $3,000 per work in the Bartz v. Anthropic settlement, which was damages for piracy, could become an informal price for training data.
• Proposed Korean rules that would require AI developers to disclose their training data are still waiting in the National Assembly.

Sources

• Seoul Economic Daily (English, AI-translated): AI Copyright Bills Stall as Google Buys Up Korean Books
• Seoul Economic Daily (Korean original): 구글도 韓 책 구매한다는데…AI 학습 저작권 강화 법안 ‘낮잠’
• Bartz v. Anthropic, order granting final approval of class action settlement (N.D. Cal., Dkt 680, 20 July 2026)
• Anthropic Copyright Settlement website

Related stories

• Publishers’ Filing Says Microsoft Told Staff to Keep ‘Project Mango’ Secret
• Meta Removes ‘Train’ Sentence From Its New Commercial Terms After Backlash
• Not Fair Use: What the Third Circuit Just Said About Training AI on Westlaw Headnotes