Anthropic maintains two artificial intelligence models apart from Claude. They're called Mythos and Fable. Recently they stopped being available to users outside the United States. The official explanation remains vague, as usually happens. What circulates among those who follow the sector closely is that the US government is requiring Anthropic to limit these models to its citizens. It doesn't seem like an unfounded rumor. It reflects something observable in other domains.

This matters because we're not talking about a technical choice. We're talking about a political decision presented as product policy. The difference isn't just semantic. A company can reverse the first. The second demands changes that go far beyond that.

I recognize these dynamics from other contexts. When a technology reaches a certain level of power, the State doesn't ask itself whether it should regulate it. It focuses on how to do so.

The relationship between Anthropic and the US government isn't new. Earlier this year, Dario Amodei publicly discussed the company's collaboration with the Department of Defense. He accepted some things and rejected others. The lines he drew then seemed firm. What's happening now with Mythos and Fable suggests those lines are shifting.

The official argument is predictable. It invokes national security. Advanced models grant strategic advantage. Allowing unrestricted access to foreign actors would, according to that logic, amount to sharing sensitive technology. It's the same reasoning previously applied to semiconductors, satellites, and encryption. It always circles back to the same point.

What that argument leaves out is equally revealing. It doesn't define who qualifies as a potential adversary. It doesn't distinguish whether a researcher in Mexico, a student in Nigeria, or a startup in Portugal represents an actual threat or simply a geopolitical inconvenience. The label "non-US citizen" encompasses billions of people. That doesn't constitute a security measure. It looks more like a barrier.

Roman stones don't lie about this. When Rome controlled Mediterranean trade routes, it wasn't out of generosity toward the provinces. That dominance over flows of information and resources allowed it to sustain hegemony without direct occupation. There are records of how certain societies restricted access to specialized knowledge, not always due to danger but to preserve advantages. Though I'm no archaeologist, these examples illuminate patterns that keep reappearing in new forms. Today's AI restrictions follow a similar playbook. Whoever regulates access to the most advanced infrastructure sets the terms for everyone else.

Here a contradiction emerges that can't be ignored. Anthropic presents itself as an organization focused on AI safety, with a mission oriented toward benefiting all of humanity rather than a single country. Its founding narrative talks about building safe AI for everyone. If that mission is real, complying with geographic restrictions imposed by a specific government is not neutral. It clashes directly with what they claim to be. If it's just PR talk, better to admit it plainly.

What's interesting is the timing. AI models are moving from experiments to essential infrastructure. Whoever has access to the best ones gains concrete advantages in productivity, research, medicine, and education. Blocking that access based on citizenship doesn't protect security. It deliberately creates technological inequality. That gap doesn't close on its own.

Researchers have been warning about these risks for some time. The danger isn't limited to powerful models falling into the wrong hands. Equally serious is the way concentrating capabilities in a single country generates structural dependencies that take decades to correct. We've seen this with patents, financial structures, and communications infrastructure. AI accelerates the process, though the incentives remain the same.

I still don't have a clear sense of how much of this responds to actual government pressure and how much Anthropic is using that pressure as cover for decisions it already wanted to make. That ambiguity is convenient for everyone involved. The government can claim it's protecting national interests. The company can say it had no choice. Either way, the result is that users outside the United States lose access.

This isn't an isolated case. It reflects a trend where advanced AI becomes a strategic resource regulated by states rather than markets. That changes the rules we thought would guide this technology's development. The idea that AI would democratize knowledge runs up against very concrete incentives to prevent that from happening fully.

From Mexico, this situation reads in a particular way. Not as an abstract debate about governance, but as a direct limitation on the tools available for building, researching, and competing. There's something uncomfortable about depending on a foreign government to decide whether you deserve access to technology that was, in theory, created to benefit humanity. That discomfort is telling. It deserves to be examined.

The question that remains open isn't whether this is fair or unfair. We already know the answer to that. The real question is what kind of innovation emerges when access depends more on a passport than on the capacity to explore.

Sources:

1. Public statements by Dario Amodei on Anthropic's collaboration with the US Department of Defense (February 2026)

2. Anthropic — Usage Policy and Terms of Service (restricted geographic access, 2025-2026 versions)

3. Bureau of Industry and Security, US Department of Commerce — Export regulations for advanced technology (EAR, Export Administration Regulations)

4. Stanford HAI — "AI Index Report 2025": geographic concentration of frontier models

5. Electronic Frontier Foundation — Analysis on AI access restrictions based on citizenship criteria (2025)