OPEN LETTER: Anthropic Cut Off, China Advances — AI Is Now a Cold War Weapon

June 2026: the month AI became a matter of state

I am writing to you from June 2026, the month when it became impossible to talk about artificial intelligence without talking about geopolitics. The American government ordered Anthropic to suspend foreign nationals' access to its two flagship models — Fable 5 and Mythos 5. Export controls, which until now had targeted primarily Nvidia chips, now extend to AI models themselves. A line has been crossed, and it is irreversible.

On June 23, 2026, ETC Journal documented what many had feared for years: the United States officially treats cutting-edge AI models as dual-use technologies, subject to the same export rules as weapons systems or nuclear technologies. At the same time, Chinese firm 360 Security announced the launch of two models — Tulongfeng and Yitianzhen — presented, with no apparent irony, as both "defensive and offensive" tools.

The paradox of the technological race creating what it claims to fight

This letter is addressed to those who, in the corridors of Brussels, Washington, Tokyo and Seoul, continue to treat AI as a matter of economic regulation, copyright or personal data protection. Those concerns are legitimate. But they are secondary to the reality of June 2026: China is producing militarized AI models, American export controls are creating side effects that open alternative markets, and hundreds of thousands of Nvidia chips have already reached Chinese subsidiaries through loopholes that the Bureau of Industry and Security (BIS) attempted to close with its May 31, 2026 directive. This race has its own logic, and it is accelerating.

I write this letter with genuine concern, not with the satisfaction of a prophet of doom. I see a technology that once promised to heal the sick, accelerate science and improve education, being progressively drawn into the logic of a technological cold war. And I am not certain that the decision-makers pulling these levers fully grasp what they are closing off.

The American decision and its justifications

The order to Anthropic to suspend foreign access to Fable 5 and Mythos 5 fits within an American national security logic that is not without foundation. These models, among the most powerful commercially available in 2026, have capabilities that far exceed text or image generation: they can assist in designing complex systems, optimizing industrial and military processes, and potentially support research on sensitive technologies. The risk that hostile state actors — China, Russia, Iran — would use them for military or espionage purposes is real.

But the measure raises difficult questions. Cutting off all foreign nationals from these models also cuts off academic researchers, legitimate businesses, and allies who could have benefited from these tools for beneficial civilian applications. The question this letter poses to decision-makers is: did you measure the cost in innovation and allied cooperation of this decision, or did you simply manage the most visible risk without looking at the side effects?

The loopholes and the hundreds of thousands of chips already gone

The BIS directive of May 31, 2026 seeks to close purchases of advanced chips through Chinese foreign subsidiaries — a loophole that, according to ETC Journal, allowed hundreds of thousands of Nvidia chips to reach the Chinese technology ecosystem. This reality reveals something essential: export controls are always lagging behind market realities. By the time a directive is conceived, approved, published and enforced, economic actors and intelligence services have already found new workarounds.

Those hundreds of thousands of chips already gone now feed the data centers of the Chinese technology ecosystem. They help train the next generation of Chinese models. China, in 2026, possesses a fleet of DUV lithography machines capable of producing millions of high-end chips domestically — a capacity developed precisely in response to earlier American sanctions. By trying to slow China down, the United States accelerated its technological independence. That is the fundamental paradox of this technological cold war.

Hundreds of thousands of Nvidia chips reaching China despite export controls — that is the perfect illustration of the gap between declared policy and operational reality. Strict controls are proclaimed from Washington, and the chips arrive in Shenzhen anyway. That demands brutal intellectual honesty about the real-world effectiveness of these measures.

Tulongfeng and Yitianzhen: names that say everything

Chinese firm 360 Security — a cybersecurity specialist intimately tied to the Beijing government — launched on June 25, 2026 two AI models named Tulongfeng ("Dragon that tames the phoenix") and Yitianzhen (names evoking legendary weapons from classical Chinese culture). According to Riccardo Ghia's Silk Bulletin No. 139, these models are presented unambiguously as "defensive AND offensive" tools. This is not clumsy marketing copy: it is a doctrinal declaration.

The launch of these models as a direct response to the suspension of access to Anthropic's Fable 5 and Mythos 5 illustrates the logic of substitution crises in the technological cold war. When a Western actor closes a door, the Chinese ecosystem opens a window. Shutting Anthropic off from foreign nationals does not slow China down: it gives it an additional justification for developing its own capabilities and deploying them without the ethical or commercial constraints that restrain American companies.

GLM-5.2 and the opportunity created by Anthropic's suspension

Silk Bulletin No. 138 from June 24 also noted that the suspension of Fable 5 by Anthropic opens "a door for GLM-5.2" — the model developed by Zhipu AI, a company linked to Tsinghua University. GLM-5.2 is now positioned as an available alternative for foreign users whom the American decision has just stripped of access to Fable 5. And this model has no American export restrictions. It is available, functional, and its international deployment can proceed without the restrictions that Washington imposed on Anthropic.

This substitution is not trivial. It means that businesses, researchers and governments around the world — including in countries allied to the United States — may find themselves using Chinese AI models by default, for lack of access to American alternatives. The decision to suspend Fable 5 and Mythos 5 for foreigners could paradoxically increase the global penetration of Chinese AI models.

Washington closes Anthropic to the world, and Beijing rubs its hands. That is the scenario tech strategists should have modelled before making this decision. I am not saying action was wrong — I am saying the side effects should have been better anticipated. A national security policy that accelerates its adversary's technological hegemony is, by definition, a failed policy.

Two ecosystems, two internets, two AIs

June 2026 accelerates a trend already visible for several years: the bifurcation of the global technology world into two distinct ecosystems — one centred on the United States, the other on China. AI is no exception to this bifurcation: we now see taking shape a world where businesses, governments and citizens, depending on their geography, will have access to fundamentally different AI models, trained on different data corpora, with different value alignments, and potentially very different military or surveillance applications.

This bifurcation has consequences for countries that belong to neither the American nor the Chinese bloc — and there are many. India, Brazil, South Africa, the Gulf states, sub-Saharan Africa: all will have to choose which technological ecosystem they inhabit, with implications for their data sovereignty, diplomatic relations and economic development. This choice is no longer hypothetical — it is pressing now, in 2026.

Europe in the blind spot of this technological cold war

I ask you directly: where is Europe in this technological cold war? It has launched the AI Act, an ambitious regulatory framework focused on the protection of fundamental rights — important, necessary, but insufficient given the geostrategic urgency. It has no world-class AI champion equivalent to Anthropic, OpenAI, or GLM-5.2. It depends technologically on the United States for its cutting-edge models — and that dependence was just bluntly exposed by the suspension of foreign access to Fable 5 and Mythos 5.

Europe must decide whether it wants to remain a spectator of the Sino-American technological cold war, or whether it wants to build its own capacity in this domain. This is not merely an industrial question: it is a question of political sovereignty. A continent that depends entirely on another power's AI models — even an allied one — for its critical decision-making processes, its defence, its administration and its economy is a continent that has delegated a portion of its sovereignty without having explicitly decided to do so.

Europe is behind on AI, and it knows it. But the real problem is not the technological lag itself — it is the absence of a clear strategy for whether it wants to close that gap, or whether it accepts permanent dependency. This technological cold war will not wait for Brussels to finish its committee meetings.

AI on the battlefield: reconnaissance, targeting, decision

This letter would not be complete without addressing the direct implications for the war in Ukraine. AI already plays an operational role in this conflict — in satellite imagery reconnaissance, analysis of Russian communication flows, drone strike targeting, and potentially in coordinating Ukrainian air defences. Access to high-performance AI models has become a measurable military advantage.

If American export controls restrict Ukraine's access to advanced models — which remains an open question in the precise modalities of the Anthropic decision — this could have direct operational implications. Conversely, if Russia or its allies can access equivalent or superior Chinese models, the technological advantage that the West has provided to Ukraine since 2022 could be progressively eroded.

The cyber war and the hacked messaging applications

In a context where the SBU and the FBI exposed on June 25, 2026 the Russian hacking of Ukrainian, European and American officials' messaging applications, the question of AI in the cyber war takes on a concrete dimension. AI models can be used to accelerate and sophisticate hacking operations — generating more convincing decoys, identifying vulnerabilities faster, analysing metadata from intercepted communications. The technological cold war and the hot war in Ukraine are not separate worlds: they feed each other.

CNBC on June 22, 2026 reported that China imposed trade restrictions on American companies in response to export controls — an escalation in the technological economic war that illustrates how every decision in this domain has chain-reaction consequences. In this environment of conflicting interdependencies, the line between economic war, technological war and classic military conflict is dangerously blurring.

I am not in a position to precisely assess the impact of the Anthropic decision on Ukrainian military capabilities — that information is classified. But the principle is clear: every technology restricted from Ukraine is an opportunity for its enemies to close the gap. And in a war of attrition, every technological advantage matters.

Think through side effects before acting

I ask you — those who hold the levers of technology policy — to model the side effects before acting. The suspension of Anthropic for foreigners may be justified from an American national security standpoint — but has anyone calculated how many legitimate users, allies and partners are now being pushed toward GLM-5.2 or the next models from 360 Security? Has anyone calculated the net advantage this decision gives to the Chinese AI ecosystem?

I also ask you to coordinate policies among allies. The United States makes its decisions, sometimes without consulting European and Asian allies who bear the consequences. An AI export control policy that is not coordinated with the United Kingdom, the EU, Japan and South Korea is a policy with gaps the size of a truck convoy. The technological bifurcation will only be contained if the industrial democracies act together.

The defining issue of the decade ahead

AI is the defining technology of the coming decades. It will reshape economies, democracies, armies and international power balances in ways we cannot yet fully foresee. What is at stake in June 2026 — who controls the most advanced models, who can access them, who is denied them, who militarises them — is not a technical detail. It is a foundational geopolitical decision. And like all foundational decisions, it will be far harder to undo than it was to make.

I wrote this letter because I believe technology must remain in the service of human freedom, not state domination. And because the technological cold war unfolding right now risks producing a world where neither democracies nor authoritarian regimes truly win — but where we all lose a little of the promise that AI carried at its beginnings.

I could have ended this letter on a more optimistic note. But that would not serve you well. The technological bifurcation is real, it is deep, and it is accelerating. What we do with that fact in June 2026 will be judged by those who live with the consequences ten years from now.

Toward an international convention on military AI?

Faced with the accelerated militarisation of artificial intelligence documented in June 2026 — Anthropic suspended for foreigners, 360 Security launching explicitly offensive models, hundreds of thousands of chips circumventing export controls — the question of an international regulatory framework is pressing with growing urgency. Several voices, notably within UN institutions, are calling for an international convention on lethal autonomous weapons systems (LAWS) that would include AI models used for military purposes.

But the obstacles are formidable. The United States and China — the two powers that have the most to lose from binding international regulation on military AI — show no eagerness to negotiate such treaties. Russia, developing its own combat AI capabilities alongside the war in Ukraine, would be a barely credible signatory to any limitation agreement. Europe, with its AI Act focused on fundamental rights, is trying to establish norms but lacks the geopolitical weight to impose them on the major powers.

The risk of AI without safeguards in combat systems

The decision by 360 Security to present its models unambiguously as "defensive AND offensive" tools reflects a disturbing trend: the normalisation of offensive AI without a clear ethical or legal framework. In the war in Ukraine, autonomous drones already using AI for reconnaissance and targeting illustrate how far this normalisation can go in a real wartime context. The line between an AI that assists a human operator and one that takes lethal decisions autonomously is progressively fading in modern conflicts.

This evolution has profound implications for international humanitarian law. The Geneva Conventions, the proportionality rule, the principle of distinction between combatants and civilians — all these principles presuppose a human decision in the chain of command. An AI system that selects and engages targets autonomously — as 360 Security's offensive models could do — creates a legal and ethical grey zone that existing international law is not equipped to handle. In June 2026, this question is no longer theoretical: it is present in the arsenals of Beijing, Moscow and potentially Washington.

An offensive AI that decides on its own targets — that is the dream of every battlefield commander under pressure and the nightmare of every international humanitarian law specialist. June 2026 normalised the idea of combat AI without providing it with a legal framework. That may be the most consequential decision of this month of June.

The urgent need for a multilateral framework

The accelerated militarisation of artificial intelligence documented in June 2026 makes the question of an international regulatory framework more urgent than ever. Several voices within UN institutions are calling for an international convention on lethal autonomous weapons systems (LAWS) that would explicitly include AI models used for military purposes. The principle is simple: technologies that can autonomously select and engage targets require the same international legal guardrails as nuclear weapons or chemical agents.

But the obstacles are formidable. The United States and China — the two powers that have the most to lose from binding international regulation on military AI — show no eagerness to negotiate such treaties. Russia, developing its own combat AI capabilities in parallel with the war in Ukraine, would be a barely credible signatory to any limitation agreement. Europe, with its AI Act focused on fundamental rights, is trying to establish norms but lacks the geopolitical weight to impose them on the major powers.

The risk of autonomous lethality without a legal framework

The decision by 360 Security to present its models as "defensive AND offensive" tools without any ambiguity reflects a deeply troubling trend: the normalisation of offensive AI in the absence of a clear ethical or legal framework. In the war in Ukraine, autonomous drones already using AI for reconnaissance and targeting illustrate how far this normalisation can go in a real wartime context. The line between an AI that assists a human operator and one that takes lethal decisions autonomously is progressively fading in modern conflicts.

This evolution has profound implications for international humanitarian law. The Geneva Conventions, the proportionality rule, the principle of distinction between combatants and civilians — all these principles presuppose a human decision in the chain of command. An AI system that selects and engages targets autonomously creates a legal and ethical grey zone that existing international law is not equipped to handle. In June 2026, this question is no longer theoretical: it is present in the arsenals of Beijing, Moscow and potentially Washington.

I remain convinced that AI can be a force for good — healthcare, education, science. But June 2026 showed that developers' good intentions are not enough to prevent militarisation. Rules are needed, institutions are needed, and the international cooperation we do not yet have. The countdown has started.Neither Washington nor Beijing nor Moscow will accept binding rules on military AI in June 2026. That is the uncomfortable truth. Which means the norms will be written not in conference rooms but on battlefields — in Ukraine today, and in the next conflict tomorrow. That is the price of this absence of international cooperation.

Acting before the bifurcation becomes irreversible

June 2026 has sounded a clear alarm: AI is now a cold war weapon. Anthropic cut off from foreigners, hundreds of thousands of chips already in China, Tulongfeng and Yitianzhen launched as offensive tools, GLM-5.2 lying in wait to capture abandoned markets — the sequence is logical, predictable, and dangerous. The technological bifurcation is no longer a distant horizon: it is here, in June 2026, documented, quantified, and accelerating.

There are still windows of action open for decision-makers willing to act. Those windows are narrow and closing. Every Chinese model deployed in a third country, every chip that circumvents controls, every ally pushed toward the Chinese AI ecosystem for lack of an available Western alternative — all of this is one more piece of the puzzle of a bifurcated technological world in which liberal democracies will have less influence. Act now, coordinate now, invest now — or accept the consequences later.

The final message

I end this letter where I began it: AI is no longer a technical issue. It is an issue of freedom, sovereignty and security. And in a week when China launches offensive AI models while the United States cuts off its tools from allies, and while Russia strikes thousand-year-old cathedrals and hacks the messaging applications of democratic leaders — in that week, we understand that all these issues are linked. Defending freedom in 2026 runs as much through firewalls and AI models as through missiles and tanks.

By Maxime Marquette, columnist

Editorial stance

I am not a journalist but a columnist and analyst. My expertise lies in observing and analysing the geopolitical, economic and strategic dynamics that shape our world. My work is to take political strategies apart, to understand global economic movements, to put the decisions of international actors in context and to offer analytical perspectives on the changes that are redefining our societies.

I do not claim the cold objectivity of traditional journalism, which stops at reporting facts. I claim analytical clarity, rigorous interpretation and a deep understanding of the complex issues that concern us all. My role is to give meaning to the facts, to place them in their historical and strategic context, and to offer a critical reading of events.

Method and sources

This text keeps the basic distinction between verified facts and interpretive analysis. The rule never changes: a factual claim is published only when a verifiable source supports it, and the sources this article actually used are listed under Sources, never here.

Primary source families used by the publication, when the story has them: official statements from governments and international institutions, public statements by political leaders, reports from intergovernmental organisations, dispatches from recognised international news agencies.

Secondary source families: specialised publications, internationally recognised news media, analyses from established research institutions, reports from industry organisations.

When an article cites statistical, economic or geopolitical data, it comes from the institutions that produce it (intergovernmental organisations, central banks, national statistics offices), and the exact institution is named under Sources.

Nature of the analysis

The analyses, interpretations and perspectives in the analytical sections of this article are a critical, contextual synthesis based on the information available, the trends observed and the expert comments cited in the sources consulted.

My role is to interpret these facts, to set them in the context of today's geopolitical and economic dynamics, and to give them a coherent meaning within the larger story of the changes shaping our time. These analyses reflect expertise built through continuous observation of international affairs and an understanding of the strategic mechanisms that drive global actors.

This article describes a documented state of affairs on its publication date, not a prediction: later developments may shift its outlook. No update is promised in advance; when an article is corrected or completed, the change is dated in the text.