On Friday 12 June, Anthropic disabled its two most capable models, Fable 5 and Mythos 5, for every customer worldwide. The company had received a US government directive to suspend access for any foreign national, inside or outside the United States (including foreign employees working inside Anthropic’s US office).
With no reliable way to separate foreign nationals from US persons across hundreds of millions of users on same-day notice, the only way to comply was to turn the models off for everyone - US and European commercial customers alike.
The security bureaucracy cited vague national security concerns regarding a potential “jailbreak” method related to code review capabilities. Anthropic says it received only verbal notice of a “potential narrow, non-universal” issue with no specific national-security detail in the letter.
Anthropic disagreed with the order, stated that the safety measures involved had been extensively tested, but had to comply anyway.
The mechanism is now on the record: a single government can issue a same-day directive and disrupt access to a globally critical AI service.
What this means for your business
Private enterprises are now caught in the crossfire of political disputes between US tech vendors and Washington’s defense bureaucracy. You have no say, no vote, and no legal recourse when these decisions are made.
This incident proves that “the cloud” is ultimately bound by geography and national jurisdiction. Because the underlying technology is governed by US jurisdiction, the US government holds a unilateral kill switch over global commercial operations.
The incident also highlights a growing divergence between regulatory approaches. European organisations are moving toward a predictable framework through the EU AI Act, while the US relies more heavily on national security interventions. Operating a predictable business that depends on both systems is becoming increasingly difficult.
It all comes down to sovereignty
This is the sovereignty argument that has worked its way up the stack for a decade, arriving now at the model layer.
It began with data residency. Choosing a European region put data in Frankfurt, but the US CLOUD Act still reached it, because jurisdiction follows the legal entity that controls the keys, not the building where it sits. It moved to silicon, where a fabrication plant on European soil remains subject to US export rules if it depends on US design tools. More recently, it reached the agentic control plane, where the gateway every AI agent passes through is, by default, a managed service inside a hyperscaler.
The model is the highest and most expensive layer of that stack. Yet, until last week, it was the one most enterprises thought about the least.
An organisation can self-host its databases, operate its own Kubernetes platform, and run its own agent gateway, yet still depend entirely on a model it can access only through a foreign-jurisdiction API. The Fable 5 shutdown showed what that dependency costs when the jurisdiction decides to act.
The decision in front of you
Ask yourself a simple question: if a foreign supplier stopped serving your organisation tomorrow morning, could you keep operating?
For many AI deployments, the answer is no.
The Anthropic shutdown established something important: the off switch exists. If a government can withdraw access to a critical service without warning, then that dependency is a business risk.
The response is to regain control.
Map the parts of your AI stack that depend on providers, gateways, or models you cannot run yourself. Then build a sovereign architecture around them. That means: infrastructure you operate, under laws you understand, with no external party able to switch it off.
Sovereignty should be treated the same way organisations already treat security and resilience: as an architectural requirement.
This is the foundation we help organisations build at Kubermatic. Our platform enables enterprises to run and manage AI workloads on infrastructure they control across on-premises, private cloud, and hybrid environments, reducing dependence on foreign-controlled platforms and services.
If sovereign AI is on your roadmap, our Sovereignty Solution Brief explains the practical steps required to build and operate AI infrastructure that remains under your organisation’s control.
