AI as a Strategic Weapon: When a Model Becomes More Dangerous Than a Missile
The Trump administration has signed an executive order treating advanced artificial intelligence models as military technologies. While the general public uses AI to write emails, governments are already classifying it as a geopolitical threat.
There is a precise moment when a technology stops being a tool and becomes a weapon. For the atomic bomb, it was July 16, 1945, at Trinity, New Mexico. For artificial intelligence, that moment may have already arrived, silently, without any visible atomic mushroom on the horizon, but with the signature of an executive order at the White House.
On June 3, 2026, Donald Trump signed a new executive order dedicated to cybersecurity and artificial intelligence. The official narrative is what one would expect: protecting American infrastructure, ensuring national security, maintaining the United States' technological primacy. But beneath the surface of this institutional communication, something far more radical is moving, something that most commentators have struggled to bring into sharp focus.
The heart of the provision concerns the most advanced artificial intelligence models, those that research laboratories are developing and that are about to be released to the public. According to what is established by the executive order, these laboratories may share their models with the federal government up to thirty days before the official release. It is not mandatory, at least formally. The administration has clarified that this is a voluntary measure, also to placate criticism received from multiple fronts. But the substance of the message is unambiguous: Washington wants to see what is about to come out, and it wants to see it before it comes out.
To understand why this is extraordinary, one must understand what is really being discussed. Not a startup testing a virtual assistant for booking restaurants. We are talking about systems like the models of Anthropic's Mythos family, which according to sources close to the cybersecurity community would already be capable of autonomously identifying zero-day vulnerabilities, meaning flaws in computer systems never previously discovered, writing functional attack code, and automating offensive campaigns against complex infrastructure. Power plants, hospitals, banking networks, government systems. All without significant human oversight. (Source: White House Executive Order on AI and Cybersecurity, June 2026; Anthropic Policy Documentation 2025.)
This is where the narrative becomes complicated, because what the American administration is doing is not simply preventive regulation. It is something deeper: it is the implicit acknowledgment that a sufficiently advanced artificial intelligence model is, for all intents and purposes, a dual-use technology, exactly like enriched uranium or advanced guided missile systems. Tools that can build or destroy, defend or attack, and that for this reason cannot simply circulate freely like a phone application.
The paradox is hard to ignore. While tens of millions of people worldwide use artificial intelligence to generate presentations, answer work emails, or complete assignments, the governments of the great powers have already shifted the conversation to an entirely different level. The discussion is not about productivity or innovation. It is about deterrence, strategic control, about who has access to what offensive capability and when.
Criticism of Trump's executive order has not been absent. From academia and industry came legitimate concerns: the risk of slowing innovation, of creating a governmental control mechanism that could transform into technological censorship, of setting a dangerous precedent for freedom of research. Criticisms that the administration has tried to soften by insisting on the voluntary nature of the measure, but which remain structurally valid.
What is clear, beyond the political controversies, is that the world has crossed a threshold. The most advanced artificial intelligence models are no longer simply technological products. They are geopolitical objects. And like all geopolitical objects, they are becoming contested ground among powers, subject to controls, restrictions, and strategies that have little to do with the democratization of knowledge or universal access to information.
The question worth asking at this point is not whether AI is truly dangerous. The question is: who decides the limits of that danger, by what criteria, and with what legitimacy? Because if the American government can block for thirty days the release of an AI model deemed too powerful, the next question is inevitable: what happens if it decides not to release it at all?
Source: White House Executive Order on Artificial Intelligence and Cybersecurity (June 2026); Anthropic Policy Documentation 2025; Center for Security and Emerging Technology (CSET), Georgetown University.