AI Is the New Uranium: Why the US Is Treating Claude Like a National Asset

This is an opinion piece. It reflects my read of where things are heading, not a neutral survey of all perspectives.

---

When Google launched in China in 2006, the philosophy was clear: get in, grow the user base, deal with the politics later. When Facebook expanded across Southeast Asia, Latin America, and Africa, the strategy was the same. When Twitter became the de facto global town square, nobody in Washington seriously considered whether that was a national security decision. Technology, the doctrine went, transcends borders. Access is good. Openness wins.

That doctrine is dead.

The US government's posture toward frontier AI — and specifically toward models like Claude — represents the most significant reversal in American technology policy since the Cold War. And most of the tech industry is either not paying attention or actively pretending it isn't happening.

---

What Just Changed

Anthropic's Fable model — its most capable to date — is not available in the same way across the world. This isn't a business decision about localisation or data residency. It's a function of a broader regulatory and export control framework that treats advanced AI the same way it treats advanced semiconductors: as dual-use technology with serious national security implications.

This didn't come from nowhere. The Chips Act, the Nvidia GPU export bans to China and Russia, the executive orders on AI safety and compute governance — these were the opening moves of a doctrine that is now reaching the model layer itself.

The logic runs like this: a sufficiently capable AI system can accelerate weapons research, compromise intelligence analysis, enable mass surveillance, undermine elections, and automate influence operations. The same model that helps a developer write better code can help a state actor synthesise bioweapons or break encryption. Therefore, it cannot be freely distributed globally in the same way that a search engine or a social network can.

That logic is not entirely wrong. But the consequences of acting on it are enormous, and largely unacknowledged.

---

The Google/Facebook Era Was an Anomaly, Not a Baseline

Here is the thing most people get wrong: the era of frictionless global tech expansion was not the natural state of the world. It was a specific historical window — roughly 1995 to 2020 — during which the US government made a deliberate bet that open technology would extend American cultural and economic influence globally, while posing limited strategic risk.

That bet paid off, spectacularly. Google became the default lens through which most of the world accesses information. Facebook shaped political discourse on every continent. American software — the App Store, Android, AWS — became the operating system of the global economy.

The government let this happen because the downside risk was manageable. A foreign actor getting access to Google Search is not a meaningful threat. A foreign government using Facebook doesn't give them anything they couldn't build themselves. The technology was powerful commercially but relatively symmetric in what it enabled.

Frontier AI is not symmetric. A model that can reason, plan, persuade, write code, and synthesise scientific knowledge at a level that exceeds most human experts is categorically different from a search engine. It is, in the language of defence planners, a force multiplier. And force multipliers get treated like assets, not like products.

So the Google/Facebook doctrine of "ship globally and iterate" was never going to survive contact with AI at this capability level. The question wasn't whether the rules would change. It was when.

---

The Strategy Anthropic Is Navigating (Whether It Wants To Or Not)

Anthropic is in an uncomfortable position. The company was founded on the explicit belief that safety is the primary challenge in AI development — that the technology is genuinely dangerous if developed without care. Its Constitutional AI approach, its investment in interpretability research, its public commitments to responsible scaling — these are sincere, not marketing.

But sincerity doesn't protect you from being instrumentalised.

When the US government decides that frontier AI is a strategic asset, the companies building it don't get to opt out of that designation. Anthropic has received significant investment from Amazon and Google. It has government contracts. It exists inside the American defence and intelligence infrastructure in ways that its founders may not have anticipated when they wrote the company's mission statement.

The result is that Claude — a model explicitly designed to be helpful, harmless, and honest — is now partly a geopolitical instrument. The restrictions on where Fable can be deployed are not just about safety. They are about maintaining American advantage in what every major government now understands to be the defining technology competition of the next fifty years.

This is not a conspiracy. It is just how strategic technology works. The same thing happened to radar, to nuclear technology, to GPS, to cryptography. The US government does not let its most powerful tools freely circulate to potential adversaries. AI has crossed into that category.

---

The Fragmentation Nobody Wants to Say Out Loud

The consequence of this policy shift is the fracturing of AI into geopolitical blocs, and that fragmentation is already happening.

China has its own frontier models, developed under state direction, not available to American users. Europe is building regulatory infrastructure that will effectively require local model variants. India is developing its own capabilities. The Gulf states are investing heavily in sovereign AI. And the US is tightening restrictions on what can be exported to whom.

This is not the internet that was promised. The original vision — shared infrastructure, global access, information as a common good — is being quietly dismantled in favour of something much more like the Cold War's divided world, but with AI systems instead of nuclear stockpiles.

The tragedy is that AI's most valuable applications are inherently collaborative. Climate modelling benefits from global data. Drug discovery accelerates with international research sharing. Language models become more capable and more fair when they're trained on the full diversity of human knowledge. Fragmentation makes all of this harder.

But national security logic doesn't optimise for global benefit. It optimises for relative advantage. And when both the US and China are operating from that logic simultaneously, fragmentation is not a risk — it's the outcome.

---

The Uncomfortable Question for the AI Safety Movement

Here is the part that nobody in the AI safety community wants to engage with honestly.

For years, the argument for AI safety research was: this technology is dangerous and powerful, we need to be careful, we need oversight, we need governance. That argument succeeded. It changed public discourse, attracted serious policy attention, and helped create the regulatory moment we are now in.

But the governance that emerged is not primarily safety governance. It is primarily power governance. The restrictions on Claude's global availability are not principally designed to prevent misuse by bad actors. They are designed to preserve American strategic advantage.

The safety framing was borrowed to serve a different agenda. And the people who built that framing — researchers, ethicists, founders who genuinely believe in the mission — are now watching their intellectual work repurposed for geopolitical competition.

This is not a reason to abandon safety research. The risks are real. But it is a reason to be clear-eyed about what "AI governance" actually means when governments get involved. It means control. And control is not the same thing as safety.

---

What Happens Next

My read: this gets worse before it gets better.

The US will extend export controls further down the capability stack as models continue to improve. The threshold for what counts as "sensitive" AI will move, and it will generally move toward restricting more, not less. Allied countries will be given preferential access, creating a tiered system that resembles NATO for AI — with all of NATO's internal tensions and exclusions.

Meanwhile, the countries excluded from the top tier will accelerate their own development. The restriction doesn't stop the technology from existing. It stops it from being shared. China already has capable models. Russia has motivation. Iran has engineers. Proliferation of capability is the historical result of every technology export control regime, because knowledge cannot actually be contained — it can only be delayed.

And the delay benefits nobody except the people who were first. For a while.

The really interesting question — the one I don't think anyone has a good answer to — is whether there is an alternative. Can you actually govern AI in a way that is both genuinely safe and genuinely global? Or are those two things in fundamental tension, given how capability at the frontier translates into military and intelligence advantage?

I don't know the answer. But I am increasingly convinced that the American tech industry's comfortable assumption — that it can build powerful AI, sell it globally, and remain outside the logic of geopolitics — was always naive. The logic of geopolitics always catches up.

Claude was built to be safe. It is now also an instrument of state. Both things can be true, and living with that contradiction is what the next decade of AI development actually looks like.