They Launched the Most Powerful AI Yet. It Lasted Three Days.

On 12 June the US government export-controlled a frontier AI model. The bigger story is who gets locked out of intelligence next.

As of 13 June 2026. This is a developing story. Anthropic is disputing the directive and says it is working to restore access, so the specifics may shift after publication. The argument below does not. A fuller version of this piece lives on my blog.

A few weeks ago I wrote about the AI Class Wars. The argument was that intelligence had quietly become a metered utility. Something you rent, not something you own. Access tiered by wallet depth, controlled by whoever owns the model, the compute, and the interface.

That was a story about money.

This one is worse.

On 12 June the US government ordered Anthropic to suspend its two newest frontier models, Fable 5 and Mythos 5, for every foreign national on earth. Bloomberg reported the directive covers any foreign national, whether inside or outside the United States, including Anthropic’s own non-citizen staff. Anthropic complied within hours and disabled both models for everyone, Americans included, because it could not separate foreign nationals from the rest of its user base fast enough to do anything else.

This was not a product launch. It was an export-control order. The same legal machinery used for weapons and enriched uranium, pointed at a reasoning engine.

Most people read it as a tech story. They are missing the signal.

This isn’t about one model.

It’s about who gets to think at the frontier.

And who doesn’t.

Intelligence Is Becoming a Strategic Asset

For decades, nations competed over the same things.

Oil. Manufacturing. Energy. Semiconductors.

Now there is a new strategic resource.

Intelligence. Not human intelligence. Artificial intelligence.

The most advanced models are becoming geopolitical assets. The same way governments once controlled uranium enrichment, rare earth minerals, and advanced chip exports, we are now watching those controls land on frontier AI.

The 12 June directive wasn’t a metaphor for this shift. It was the shift, in writing, signed.

The most valuable resource of the next decade won’t be oil. It will be access to reasoning.

The AI Divide Is Replacing the Digital Divide

The internet created winners and losers.

Countries with broadband surged ahead. Those without it fell behind.

AI creates a far larger gap. Because intelligence compounds.

A developer using frontier AI is more productive. A researcher moves faster. A startup ships quicker. An entire economy becomes more competitive.

The difference between having access and not having it isn’t incremental. It’s exponential.

And unlike broadband, AI access can be switched off instantly. One letter. One directive. One signature. We just watched it happen.

Suddenly the world’s most powerful cognitive infrastructure becomes unavailable. Not because you can’t afford it. Because you are not allowed to use it.

Intelligence Is Now Export Controlled

This is no longer a forecast. The pattern is on the record.

First the US restricted advanced GPU exports. Then semiconductor manufacturing equipment. Now frontier AI models, named individually, pulled by a government letter citing national security. The stated trigger was a reported method of bypassing a safeguard that stops the model finding software vulnerabilities, a rationale Anthropic publicly disputes and calls a likely misunderstanding.

Dispute or not, the precedent is set.

Modern models aren’t chatbots. They are scientific research accelerators, cybersecurity tools, planning assistants, productivity engines, and increasingly autonomous decision systems.

Governments understand something many investors still don’t. AI isn’t another software category. It’s infrastructure. And infrastructure is always political.

The country that controls intelligence infrastructure controls economic gravity.

The Three-Tier AI Future

If this continues, the world splits into three AI classes.

Tier 1. The AI superpowers. Countries with unrestricted access to frontier models and compute. The United States. A handful of close allies. The nations that host the largest labs.

Tier 2. Licensed intelligence. Countries granted approved access under regulatory frameworks. Useful models. Limited capabilities. Controlled distribution.

Tier 3. The AI periphery. Everyone else. Dependent on older models, open-source alternatives, and delayed releases.

The result isn’t a knowledge economy.

It’s an intelligence hierarchy.

The Open-Source Counterattack

Every restriction creates pressure. And pressure creates alternatives.

If frontier AI keeps getting gated, open-source ecosystems accelerate. Not because they are better. Because they are available.

History repeats this. When access tightens, markets fork, developers adapt, and communities build around the wall.

The more intelligence is centralised, the stronger the incentive to decentralise it. This is where open models, decentralised compute, and tokenised GPU infrastructure stop being ideology and start being resilience.

The View From Outside

I’m writing this from England.

Which means under the directive I just described, I am exactly the person it locks out. Not for cost. Not for capability. For nationality.

The internet started open. The cloud concentrated. Social media centralised. AI is now running the same play, only faster. And this time a single government letter can pull a frontier model from everyone who isn’t American, overnight.

So the real question isn’t whether Anthropic restricted Claude. It didn’t choose to. A directive did, and the company is fighting it.

The real question is whether access to intelligence is becoming a privilege of passport and postcode.

Because if intelligence is gated by geography, by politics, by citizenship, then we didn’t democratise it.

We privatised it. Then we nationalised it.

And that’s a very different future.

The New Borders Are Invisible

The next great geopolitical battle won’t be fought over land.

It will be fought over intelligence. Not who owns it. Who can reach it.

In a world where AI becomes the operating system of science, business, education, and government, restricting intelligence is the same as restricting opportunity.

The internet connected the world.

AI may divide it again.

And the biggest question of the next decade isn’t “How smart will AI become?”

It’s “Who gets to use it?”