US Commerce Bars Foreigners From Anthropic's Top AI Models
The US Commerce Department has ordered Anthropic to wall off access to its most advanced models from foreign nationals, framing the directive as a necessary safeguard for national security. One notices, in the fine print of such decrees, the specific definition of “foreign national” which often excludes the very engineers who built the wall. With that detail load-bearing - where the architects of the barrier are themselves subject to the barrier the barrier is meant to protect - the announcement reads not as a fortress, but as a cage where the jailers are also inmates.
The standard framing assumes that technology, like uranium, is a static resource that can be contained within borders by sheer administrative will. This is a category error of the highest order. AI models are not physical assets; they are patterns of weights and biases, compressed into tensors that travel at the speed of light across fiber-optic cables. To restrict access to foreign nationals is to attempt to dam a river by asking the water to identify itself before it flows. The US government operates under the illusion that it can legislate the movement of information as if it were contraband gold. But information does not respect passports. It respects bandwidth.
Consider the incentive structure of the engineers at Anthropic. These are not faceless bureaucrats following orders; they are the schaap met vijf poten - the rare candidates who understand both the code and the consequence. When a directive arrives that contradicts the fundamental nature of their craft, they do not simply obey. They look for the leak. The “wall” is not a single gate; it is a thousand micro-gates, each one a potential point of failure or circumvention. The more complex the restriction, the more likely the system is to break in a way that benefits the very actors it seeks to contain. A foreign researcher in Seoul or London does not need the full model weights to understand the architecture. They need the gradient descent logic, the attention mechanisms, the loss functions. These are mathematical truths, not proprietary secrets. You cannot patent gravity.
The Commerce Department’s order reveals a deeper anxiety: the fear that leadership is no longer a function of geography, but of computation. By restricting access, the US is not securing its advantage; it is forcing the rest of the world to build their own advantage, independently of American oversight. This is the unintended consequence of every export control on software. It drives innovation underground, into the dark corners of the global internet where no regulator can reach. The “wall” becomes a catalyst for the very decentralization it seeks to prevent.
There is a specific incident from the early days of the open-source movement that illustrates this point. A company attempted to lock its core algorithm behind a strict API limit for international users. The result was not compliance, but a flurry of reverse-engineering efforts that produced a superior, faster, and completely free alternative within eighteen months. The original company lost its market share not to a competitor, but to the community it tried to exclude. The “wall” did not protect the asset; it accelerated its obsolescence.
The plain question for the room is this: if the technology cannot be contained by borders, and the people who build it cannot be controlled by decree, what is the actual function of the restriction? Is it to protect the technology, or to protect the illusion of control? The answer matters. If it is the former, the policy is doomed to fail. If it is the latter, it is a political performance that wastes the resources of the very institutions it claims to defend.
The people inside the system - the engineers, the legal teams, the compliance officers - are not the enemy. They are the ones trying to make a square peg fit into a round hole. They are doing their best with a tool that does not exist. The system, however, is broken. It is trying to apply a 20th-century industrial logic to a 21st-century digital reality. The result is not security. It is noise.
The final image is not of a wall, but of a mirror. The US government is looking at its own reflection in the glass of its own technology, and seeing a shadow that is already moving elsewhere. The wall is not keeping anyone out. It is keeping the truth in. And the truth is that the future of AI is not American, or Chinese, or European. It is global. And it is already here.