Deep Dive · XiaoHu Explains

Being able to switch models at any time is itself a form of deterrence

The US export controls of June 12 marked the first time in history that frontier AI models themselves — not just chips — were brought under export control. Ren Ito's prescription: master the art of multi-model orchestration.
30-Second Recap
  • Ren Ito, co-founder of Sakana AI and a former Japanese diplomat, published a signed op-ed in The Yomiuri Shimbun arguing that the next AI race will be won on systems capability, not model capability.
  • On June 12, 2026, the US government issued export controls that, for the first time in history, placed frontier AI models themselves under restriction — previously only AI chips were controlled — this time limiting foreign access to Anthropic's most advanced models.
  • The piece draws on two historical precedents: the US sharing cryptographic technology with allies during the Cold War, and the divergent paths of the F-35 (a multinational joint program) and the F-22 (US-exclusive) — illustrating that "share vs. hoard" has always been a strategic choice.
  • The author's core argument: AI is an ecosystem of multiple coexisting frontier models, with new members emerging at breakneck speed. The competitive question has shifted from "whose model is strongest" to "who can reliably access the strongest AI whenever needed."
  • His prescription for Japan: its strength lies in integrating technology into trustworthy systems — it should invest in independent evaluation, interoperability, and secure deployment, rather than trying to replicate every frontier model at home.
01Backdrop · After the G7

After one G7 meeting, the conversation changed

Ren Ito, co-founder of Sakana AI and a former Japanese diplomat, wrote this signed op-ed for The Yomiuri Shimbun following the G7 summit in Évian, France.

At the summit, he noticed that the question world leaders and tech leaders were asking about AI had changed. In the past, everyone cared about whose AI model was strongest. Now, what's really unsettling people is whether they'll keep having access to the most frontier AI at all.

What made this shift impossible to ignore was something that had just happened.

2026.06.12
The US government issued export controls restricting foreign access to Anthropic's most advanced AI models. In the past, export controls only targeted AI chips — this time, frontier AI models themselves were included, for the first time in history.

For the past few years, the main storyline of this AI race has been making models bigger and stronger: tech companies pouring hundreds of billions of dollars into compute, with success measured by benchmark rankings, parameter counts, and training scale. Today, competition looks different. Who gets access to frontier AI, and on what terms, is increasingly a matter decided by geopolitics.

02Historical Precedent

Historically, "share" vs. "hoard" has always been a choice

Ren Ito says there's historical precedent for how to handle a leading technology. During the Cold War, the US repeatedly faced the same choice: should an advanced technology be shared with allies, or kept close?

He cites two sets of examples. During the Cold War, the US shared advanced cryptographic technology with trusted allies, because secure communications made the entire alliance more resilient. Decades later, the same logic applied to fighter jets — just split across two different paths.