A Blockade on
Knowledge. After June
12, Nothing Will Be the
Same
The most advanced artificial intelligence model was switched off for everyone who is not
American. What used to belong to political fiction became fact on a single Friday evening, and it is
a signal that neither Europe nor China has any right to ignore.
On June 12, 2026, the United States government, citing national security, ordered the company
Anthropic to cut off access to its most powerful models, Fable 5 and Mythos 5, for anyone who is
not a US citizen, inside the country and beyond it. The company could not carry out this order
directly, because a significant part of its own staff are not Americans. The only way to comply with
the directive was to disable the model for everyone. Fable 5 stopped working two days after its
release.
We cannot wait for an American AGI for everyone, because it will not come. They will not make it
available to the world. They will make it available to their own citizens, and perhaps to a few
satellites. The rest will get access to stripped-down, supervised, severed versions, or nothing at all.
And this is not because the United States is evil. It is because any state that holds such an advantage
will, sooner or later, use it. This is not a question of morality. It is a question of the logic of power.
American elites have spoken for decades about exceptionalism and leadership. They have now
shown what that means in practice: we have the best tool, therefore you cannot have it.
Can we then harbor any illusion that, were an AGI to emerge, the US would not block access to it?
We cannot. Not for a moment. They will block it, immediately, and there will be no appeal. Whether
Fable 5 was a deliberate test or the reflex of a panicked administration is secondary. What matters is
that it functioned as a test, and the result is unambiguous: it proved feasible, unpunished, accepted
by the world without resistance. Now it is known that the same can be done with every model that
follows. Every one. AGI included.
This is no technical formality, nor an isolated incident. It is the first known case in which a federal
government forced a leading artificial intelligence company to withdraw a publicly deployed model.
The precedent has been set, and precedents get repeated.The official reason was a narrow, as it was put, vulnerability, a way of bypassing the model’s
safeguards. Anthropic itself publicly disagreed with the decision. It stated that the same
vulnerability is probably present in other publicly available models, including OpenAI’s GPT-5.5,
that the evidence had been conveyed only verbally, and that if such a standard were applied across
the industry, it would halt the deployment of all new models. The stated reason therefore does not
account for the scale of the response.
And the logic of the decision is this. Had the danger lain in the tool itself, it would have been cut off
for everyone and in all of its forms. Instead, the same capability remained available in competitors’
models, which no one switched off. The criterion for cutting access was not “dangerous or safe.” It
was “ours or foreign.” And a tool that some may use and others may not, by passport rather than by
risk, is not being restricted for safety. It is being restricted for advantage.
This is why neither Europe nor China can hold any illusions any longer. The message reads: the
most powerful cognitive tools of the age will be granted to one’s own and denied to outsiders, and a
single capital will decide who counts as one’s own. Anyone building their future on the assumption
of permanent access to American models has just seen what that assumption is worth: two days.
And here we reach the heart of the matter, which extends beyond this one decision. The knowledge
that is the raw material of artificial intelligence did not originate in the United States, and did not
originate in any single country. It was built over generations, on every continent, in every language
that has ever recorded a thought. Models are trained on that inheritance in its entirety, because
otherwise they would not be what they are. What we call artificial intelligence is the condensed
effort of humanity, which no single party produced and no single party owns. Newton wrote that he
saw further because he stood on the shoulders of giants. Every AI model stands on the shoulders of
all the giants at once, and that is precisely why it cannot be anyone’s national property.
A company that makes such a model available is not its owner in the sense in which one owns a
thing made from scratch. It is a custodian. An administrator of a common good belonging to others,
charging a toll, a subscription, for converting electricity into an answer. This is a fair exchange and
no one disputes it: maintaining the infrastructure costs money, so a fee for access is just. But a fee
for access to the deposit is not the same as title to what lies within the deposit. And what lies within
the deposit is the knowledge of humanity. Still less does title belong to a government that produced
neither the knowledge nor the model, and merely issued an order.
Someone will say that knowledge is not like water or air, which are common by necessity, whereas
access to a powerful tool can be rationed. But knowledge is a good even harder to appropriate than
water, not easier. Water once drunk is gone. Knowledge once passed on does not diminish, it
multiplies: when I know, you do not know less. There is no scarcity that would justify rationing it.
Cutting someone off from knowledge protects no dwindling resource. It protects only the advantage
of the one doing the cutting.
Knowledge is the property of all humankind. Not because some government decreed it, but because
no government created it. One may charge a toll for electricity turned into thought. One may not
blockade access to knowledge itself according to whose citizen one happens to be.
And this is precisely why the conclusion from June 12 is not “America is the enemy.” It is different,
and graver. If this is the logic of power, and not the ill will of one country, then we must not make
our own cognitive future dependent on any single power, because each of them, sooner or later, will
act the same way. Cognitive sovereignty, our own models, our own infrastructure, our own access tothe inheritance of humanity, is not a luxury or an ambition for show. It is a condition of
independence in an age in which thought is assisted by machine. Waiting on another’s grace ended
on June 12. The reckoning on ourselves began