On the afternoon of June 16, I was sitting in a banquet hall at Conrad Seoul in Yeouido.

It was Claude Build Day Seoul, hosted by Anthropic together with Bass Ventures. Each of the 22 tables had its own mentor, which made the event feel less like a one-way lecture and more like a dense, focused conversation. At our table, the CTO of MyRealTrip shared the company’s experience with AX, or AI transformation. One story stayed with me: a non-technical team member, someone without a development background, had used public APIs to produce meaningful results in a short period of time.

The message that ran through the room that day was clear.

The real value of AI transformation is not simply having better tools. It is lowering the barrier so that more people can actually use them.

And yet, I was listening to that optimism under a strange shadow.

Just a few days earlier, the U.S. government had ordered Anthropic to block two of its most powerful AI models from all foreign nationals. Those of us living in Busan are included in that category. In one room, people were saying, “Now anyone can build.” In the same week, on the other side of the world, a decision was made that said access could be cut off based on nationality.

The two scenes sat uneasily beside each other.

Here is what happened.

On Friday, June 12, 2026, at 5:21 p.m. Eastern Time, Anthropic received a U.S. export control order based on national security concerns. The order was simple in its wording but enormous in its implications: Claude Fable 5 and Claude Mythos 5 had to be made unavailable to foreign nationals. This applied whether those people were inside or outside the United States. It even included foreign-national employees working at Anthropic.

Because the line was drawn by nationality, the company effectively had to shut off access to the two models for nearly all customers. Over that weekend, two of the most advanced AI models in the world went dark.

Anthropic publicly stated that it disagreed with the measure. The government’s concern was reportedly tied to the possibility that Claude Fable 5’s safeguards could be bypassed to unlock stronger cybersecurity capabilities through a jailbreak. Anthropic argued that this was a narrow case, not a general pattern, and that similar issues could also be found in other publicly available models. The company warned that if one narrow vulnerability were enough to force the withdrawal of a commercial model used by hundreds of millions of people, the release of all frontier models could effectively grind to a halt.

On the other hand, the government’s argument was grounded in national security and cyber vulnerability detection. There is also a certain irony here: Anthropic itself has often argued for restricting access in the name of AI safety. In this clash, both sides have points worth taking seriously.

Fortunately, other models, including Claude Opus 4.8, were not affected.

That is the factual outline.

Now let us look at the light and shadow of this event from the perspective of someone with South Korean nationality, living in Busan.

First, the shadow.

The incident made one thing painfully clear: the power switch for frontier AI is in the hands of geopolitics. A tool that worked perfectly yesterday can be turned off overnight, not because of anything you did, but because of a decision you had no vote in. If a startup in Busan had built its product on top of Fable 5, that company could have faced sudden disruption without having done anything wrong.

Second, the line was not drawn based on what someone had done. It was drawn based on where someone belonged. Koreans are clearly foreign nationals under this rule. For the first time, we saw with painful clarity a situation where access was divided not by ability, trust, or behavior, but by nationality.

Third, the impact spread far beyond the original target. The order was aimed at foreign nationals, but separating access cleanly by nationality proved difficult in practice. As a result, even U.S. customers lost access to the models. It showed how fragile it can be to build an entire stack on a single frontier AI provider.

Fourth, the precedent is frightening. If this kind of restriction becomes normal, the most advanced layers of AI may increasingly be locked behind nationality and geography. For non-U.S. and non-capital-region ecosystems that hope to build their AX future on borrowed frontier capabilities, this is a structural headwind.

But the story is not only negative.

There are also opportunities left for Busan.

First, there is the question of proportion. What went dark was only the very top frontier tier: Fable 5 and Mythos 5. Most of the work happening in real-world settings still relies on models such as Opus 4.8 and other available systems, which remained intact. The headline sounded dramatic, but for many builders in Busan, the work they were doing yesterday could still continue today. The practical break was much narrower than the title suggested.

Second, this incident gives us a vivid reason to talk seriously about AI sovereignty and resilience. Do not bet everything on one model. Do not depend entirely on one provider. Design alternatives and fallback paths before a crisis arrives. For Busan, Ulsan, and Gyeongnam, especially if manufacturing AX is part of the region’s future, this is a powerful moment to discuss resilience in a concrete way.

Third, after the shock, one thing becomes clear: what matters is not just a single API key, but relationships and the ability to rearrange quickly. The value of trusted local hubs increases. The value of long-term partners with real authority in the region increases. If Busan can become not just a passive consumer of AI, but a node that helps people navigate instability, then this kind of crisis can become both a warning and an opportunity.

Finally, the global movement toward a more distributed AI landscape may contain favorable winds for Korea and Busan. Whether we remain passive users or rise into serious regional hubs will depend on the choices we make now.

The optimism that filled Claude Build Day in Yeouido — the belief that anyone can now build — becomes truly meaningful only when that “anyone” is not standing somewhere that can be switched off with a single command.

For someone living in Busan, the real question raised by this incident is not, “Which model is the strongest?”

The question is this:

When the switch is turned off, how solid is my position, and how deeply am I connected?

This block is both a warning and an invitation. It warns us about the fragility of borrowed power. At the same time, it invites Busan to build a place of its own — one that does not collapse the moment the switch goes dark.