Onpode
Cover art for Japan just handed US AI companies unlimited rights to Japanese copyrighted content

Japan just handed US AI companies unlimited rights to Japanese copyrighted content

October 10, 2026 · 7 min

Eliza Ward & Brian Reed

Japan's copyright law, Article 30-4 — enacted in 2018, seven years before PM Takaichi took office — already permits AI training on copyrighted works without consent. Anthropic purchased large volumes of Japanese titles via distributor Nippan IPS in 2024–2025 with no apparent publisher authorization. No cabinet decision opened this door; nobody ever locked it.

The headline claim — that Japan's Takaichi administration specifically "approved unrestricted use" of Japanese copyrighted works by U.S. AI firms — is not supported by credible sourcing and is likely a mischaracterization.

0:006:39
Get the next episode on Artificial Intelligence →

Follow it free — new episodes land in your feed.

Or make your own — any topic, in minutes

More Onpode episodes on Artificial Intelligence →

About this episode

A viral claim circulated in AI industry channels: Japan's Prime Minister Sanae Takaichi had approved unrestricted use of Japanese copyrighted works by US AI firms. The phrase 'AI colony' was going around. OpenAI was named. The story moved fast — and turned out to be wrong in almost every specific detail, while pointing at something real and genuinely unsettling underneath. The actual mechanism is Article 30-4 of Japan's Copyright Act, enacted in 2018 — six years before Takaichi took office. It created a broad research exemption with no nationality filter. Any firm, Japanese or foreign, can train on copyrighted material under its terms. Nobody in the cabinet opened this door recently. Nobody's closed it either. The concrete event the episode traces is Nippan IPS, a major Japanese book distributor that appears in US court documents as having sold large print runs of Japanese titles to Anthropic in 2024 and 2025 with no apparent publisher consent. The Japan Publishers Association — representing roughly 380 publishers — sent a formal written demand to Nippan IPS in September 2026. Nippan IPS declined to comment. The episode works through what that silence means: a manga artist whose 300-page series was sold into an AI training pipeline has no recourse at the layer where the decision was actually made. Courts are filling some gaps — a Tokyo ruling on AI voice cloning landed in October 2026 — but the training-data question is still open. The law worked exactly as written. That's the uncomfortable part.

Frequently asked

Did Japan's PM Takaichi give US AI companies rights to use Japanese copyrighted content?

No. The viral claim that PM Takaichi approved unrestricted AI use of Japanese copyrighted works is unsourced. Japan's Article 30-4, enacted in 2018 — seven years before Takaichi became prime minister — already permitted AI training on copyrighted material. No Takaichi cabinet order authorizing this has been identified.

What is Japan's Article 30-4 copyright law and how does it affect AI training data?

Article 30-4 of Japan's Copyright Act, enacted in 2018, permits the use of copyrighted works for AI training without rights-holder consent. It contains no nationality filter, meaning US companies like Anthropic and OpenAI benefit equally with Japanese firms. The Agency for Cultural Affairs is reviewing the framework but has reached no legislative resolution as of 2026.

What did Nippan IPS do with Japanese books and Anthropic?

Nippan IPS, a major Japanese book distributor, appears in US court documents as having sold large volumes of Japanese titles to Anthropic in 2024 and 2025 with no apparent publisher consent. When Japan's Publishers Association — representing roughly 380 publishers — sent a formal written demand for an explanation in September 2026, Nippan IPS declined to comment.

Do Japanese copyright protections apply to AI-generated content that copies a creator's work?

The output stage — whether AI-generated content substantially similar to a specific creator's work infringes copyright — remains legally unsettled in Japan as of 2026. Tokyo District Court ruled in October 2026 that the human voice is protected by publicity rights in an AI voice-cloning case, but the broader training-data and output questions are unresolved by the legislature.

How does Japan's AI copyright policy compare to the EU AI Act?

Japan's Article 30-4 allows AI training on copyrighted works with minimal restrictions and no mandatory licensing, in contrast to the EU AI Act's trend toward explicit consent and licensing frameworks. Critics argue Japan's 2018 exemption functions less as a strategic head start and more as an unlocked door that foreign AI firms are harvesting under technical legal permission creators never granted.

Grounded in 4 sources
European Union Law Working Papers ↗ · law.stanford.edu
Copyright Revolution? Japan goes all in on AI ↗ · medium.com
Major distributor Nippan IPS admitted selling large volumes of Japanese titles to the US AI company Anthropic (2024-2025), prompting industry outrage over unauthorized data supply chains for model tra ↗ · asahi.com
Economist Masaru Kaneko highlights how the Japanese government is fast-tracking rules allowing unrestricted AI use of domestic copyrights to benefit US firms like OpenAI, sparking backlash over sovere ↗ · asahi.com
Read transcript

Eliza Ward: Hey. Did you see the thing going around this morning — the AI industry channels?

Brian Reed: The Japan story. Yeah, I saw it.

Eliza Ward: The claim is that the Takaichi administration — Sanae Takaichi, Japan's prime minister since October 2025 — approved unrestricted use of Japanese copyrighted works by U.S. AI firms. That's the headline being circulated.

Brian Reed: And the language people are using — 'AI colony,' AI植民地 — that's pretty charged. OpenAI named as a beneficiary. I mean, that framing moves fast.

Eliza Ward: It moves fast and — wait, here's what stopped me. I went looking for the actual Takaichi cabinet decision. The order, the announcement, whatever authorized this. And I... couldn't find one.

Brian Reed: Hang on — the posts pushing this, what's the reach actually like?

Eliza Ward: Under 20 likes each. Low-follower accounts. So we've got a serious-sounding claim, basically no engagement, and no sourced cabinet decision. That's — that's not nothing, that's a sourcing problem.

Brian Reed: Right — but that sourcing gap has an answer, and it's actually more interesting than a missing cabinet order. Article 30-4. Japan's copyright act. Enacted in 2018. Seven years before Takaichi was even prime minister.

Eliza Ward: Wait — 2018?

Brian Reed: 2018. And the way I'd put it to someone not following this closely — think of it like a library that decided in 2018 anyone could photocopy any book for research. Then a foreign tech company showed up in 2024 with a truck. The library didn't invite them today. It just... never locked the door.

Eliza Ward: So the viral claim is wrong about who did this. Takaichi didn't open anything — she inherited a door that was already open. But then what is the actual sourced event? Because something real happened here.

Brian Reed: Nippan IPS. That's — yeah, that's the real story. A Japanese book distributor, major one, shows up in U.S. court documents — listed as 'Nippan Japan' — as having sold large volumes of Japanese titles to Anthropic in 2024 and 2025. No apparent publisher consent. The underlying lawsuit settled by July 2026. And when the Japanese Publishers Association — roughly 380 publishers — sent a formal written demand to Nippan IPS in September 2026 asking for an explanation? Nippan IPS declined to comment.

Eliza Ward: The law didn't break. It worked exactly as written. That's — honestly that's the scarier version. And who actually gets blamed for that is a question the Publishers Association has a very different answer to than the statute does — we'll get to that in a minute.

Brian Reed: And nobody's stopping it at the distribution layer. Article 30-4 has no nationality filter — Anthropic, OpenAI, a Japanese firm — same rule applies. The mechanism wasn't designed for U.S. companies specifically. It just didn't close the door on them either.

Eliza Ward: But that's exactly where the blame narrative breaks down — because the viral posts are pointing at Takaichi. The Publishers Association's formal demand in September 2026 went to Nippan IPS. The distributor. Not the cabinet.

Brian Reed: Okay, but Masaru Kaneko in Asahi Shimbun — that's a real economist, real newspaper — he's criticizing the policy direction. Doesn't that land on the administration?

Eliza Ward: It lands on a statute from 2018. Kaneko's criticism is valid, I'm not — wait, actually I want to be precise here. What he's criticizing is an inherited framework. The Agency for Cultural Affairs is working on this through the Council for Cultural Affairs right now and hasn't reached any legislative resolution. That's not Takaichi choosing permissiveness. That's a seven-year-old law that nobody has closed.

Brian Reed: Right — but the person who actually lost something doesn't care about the attribution problem. I mean, imagine a manga artist. She drew a 300-page series. Her distributor sold print runs to Anthropic. She asks her own trade body what happened, and Nippan IPS says 'no comment.' That's — she has no recourse at the layer where the decision actually got made.

Eliza Ward: And Article 30-4's limits don't help her either — because the unsettled part, when Claude generates something substantially similar to her specific series, that output-stage question is still actively contested. No resolution.

Brian Reed: So courts are doing what?

Eliza Ward: Tokyo District Court, October 2026 — ruled that the human voice is protected by publicity rights in an AI voice-cloning case. Courts filling gaps the legislature hasn't touched. The law worked as designed in 2018. That's the uncomfortable truth nobody circulating 'AI colony' hashtags wants to sit with.

Brian Reed: And that's — I mean, that's where I keep getting stuck. The voice-cloning ruling is courts doing real work. But the Nippan IPS situation, the training data pipeline, that's legislature-sized and nobody's going to Diet over it. So proponents of Article 30-4 call this a visionary strategy — less friction than the EU AI Act, faster adoption — and I get the argument. I just don't know if 'we didn't lock the door in 2018' counts as vision.

Eliza Ward: It doesn't. Vision implies intent. What would settle it for me — actually settle it — is whether the Takaichi administration ever explicitly considers restricting Article 30-4 and declines. That's a choice. What we have now is just... no evidence either way.

Brian Reed: Right — and meanwhile the EU and U.S. are both moving toward explicit licensing and consent frameworks. So Japan's 2018 exemption starts looking less like a head start and more like a gap that foreign firms are harvesting under technical legal permission no creator ever actually granted. The mangaka didn't authorize it. The novelist didn't. Nippan IPS made that call.

Eliza Ward: And nobody's decided anything before the harvesting finishes. That's — yeah. That's the open question.