Nippan Group Holdings, Japan’s largest book distributor, has reportedly sold large quantities of books to US AI developer Anthropic — and it has now emerged that the Japan Book Publishers Association (JBPA), which represents roughly 380 publishing houses, sent Nippan a written request demanding it confirm the facts. The books that changed hands may have been used to train AI, raising questions about the legal legitimacy of selling them in bulk to an overseas company without publishers’ consent. Publishers have long voiced concerns about AI training on their works, and the letter also demanded that Nippan clarify its policy going forward. On 5ch, opinions split not only over whether the sale itself was appropriate, but also over the relationship between AI training and copyright, and whether creative work will ultimately be replaced by AI.
It emerged on the 18th that the Japan Book Publishers Association (JBPA), an industry body representing roughly 380 Japanese publishers, sent major book distributor Nippan Group Holdings a written request to confirm facts, after reports that Nippan had sold large quantities of books to US artificial intelligence developer Anthropic. Publishers have pushed back, citing the possibility that the books sold were used by Anthropic for AI training.
According to several people connected with the publishing industry, JBPA’s letter asked whether the books that went to Anthropic included titles published by its member publishers, and if so, for details on those titles. It also asked about the legal legitimacy of Nippan selling large quantities of books to an overseas company without publishers’ consent.
Publishers have long been concerned about their works being used to train AI. They called on Nippan to halt any new sales of books intended for AI training and to lay out its stance on supplying books for that purpose.
Source: asahi.com / Read the original article here
I work at a foreign-affiliated law firm, and it's replacing a lot of the work already.
We're planning to lay off several staff whose jobs got taken by AI agents by the end of the year.
Overseas news has also reported that AI service companies are buying up huge quantities of old books. Sounds like they're taking apart valuable antique books to scan them.
Does the physical book format really matter? What matters is the content written inside. They're digitizing it and sharing it with the world — we should be thanking them, not complaining.
Guess that's what happens once online sources alone aren't enough anymore…
Yeah, there's no reason or legal basis to refuse the sale.
If they didn't get the publishers' consent, the distributor's just a middleman pocketing a cut.
What exactly would that violate? Is it actually prohibited?
US TV shows and movies have gotten seriously boring lately. Writers have been on strike forever over AI disputes. Seems there's been some compromise, so things have eased up a bit. But AI can't do creative work. That'll become even clearer over time.
Whether it's 'impossible'? Some people say it's just that the training isn't sufficient yet. I think so too. I mean, it's not producing Nobel-Prize-level highbrow literature, but it can already create stuff good enough to eat up ordinary people's free time, so in a sense the groundwork for making it a viable business is already there.
What you're seeing in American dramas is just the tiny fraction that succeeded. Because human output is limited, you were only ever seeing the top 100 best works — going forward you'll be seeing the top 1,000,000 best works. A hit drama from 30 years ago wouldn't be nearly as popular if you watched it today. If it had a 30% rating back then, would it still get 30% now? Same thing — the pace of change is just going to accelerate.
It's strong at statistics, but apparently it still can't even beat humans at finding the optimal setup for Mini 4WD (a Japanese toy race car).
Translation's the main use case, right? AI translation accuracy keeps improving.
The fact that tons of people are happy with that level of quality and it's already turning a profit means it's already practical. That's good enough for now.
Chinese AI short dramas — AI-made short dramas — are on track to become mainstream. Tons of them are up on YouTube too, and if you watch, you'll notice they have a distinctive quirk to them.
You should go read up on AI and copyright law. What Anthropic did here is legal. The publishers biting back is nothing but emotional argument.
You can't use something commercially when the copyright holder has no idea it's happening.
Anthropic hasn't paid copyright holders a single yen, and it hasn't been granted permission for commercial use by them either. What part of that is legal?
Anthropic actually bought the books, you know? Authors get royalties from that. They did pay for them. And copyright law explicitly states that no permission is needed for AI training use. If you don't like that, you'd have to change the law in both Japan and the US.
If you scanned someone else's copyrighted work yourself (jisui, DIY book-scanning) and published it, that'd be a crime. There's value in nipping that kind of thing in the bud.
Anthropic is doing commercial use, though, so that doesn't fit the same definition.
Well, AI output's full of errors anyway, so who cares, right?
The accuracy's actually good these days — if you have an AI read it, that is. Run-of-the-mill OCR software won't cut it.
So could I get sued too, then? Preemptively, in case I post a book I bought straight onto the internet? lol
If Anthropic ever actually starts publishing the full contents of the books it bought, that's when you sue.
You'd act before the damage happens, right? You'd act to protect your interests.
Well, I think it's fine to voice your worries and act on them — that's exactly what this news story is, I guess. Freedom of thought and speech and all that.
It's just that it won't have any real effect.
Why would you assume Japan alone gets a pass? In the US, Anthropic was forced to pay around $1.5 billion in a class-action copyright infringement lawsuit over AI training.
I support the US's stance on copyright protection too, but could Japan actually produce the same kind of ruling? There's not even a court decision yet on image-scraping either.
Given where Anthropic is now, they'd probably just cough it up as a necessary cost of doing business.
Background and Key Points of Debate
Article 30-4 of Japan’s Copyright Act permits the use of copyrighted works for “information analysis” purposes such as AI training without the rights holder’s permission, so buying books, scanning them, and turning them into training data isn’t automatically illegal under domestic law. In the US, however, Anthropic settled with authors for around $1.5 billion in 2025 over using pirated books to train its AI, and within the thread this precedent became a basis for the more cautious view that “simply paying for the books doesn’t settle the matter.” The debate splits broadly into two points: first, whether it was appropriate for Nippan to sell books in bulk without obtaining consent from individual publishers, and second, even if training-purpose use is legally permitted, whether routing it through a distributor’s supply chain is appropriate under publishing contracts — a question that can’t be resolved by copyright law interpretation alone.
*This article was compiled by excerpting and summarizing the 5ch (News Speed+) thread “Book Distributor Nippan Allegedly Sold Massive Numbers of Books to US AI Company — Industry Group Demands Explanation.”
