When Compute Can Be Switched Off

A US export directive forced Anthropic to disable its most advanced models worldwide, turning access to frontier compute into a question of national strategy.

What Happened

On June 12, the United States government issued an export control directive instructing Anthropic to suspend access to its two most advanced systems, Fable 5 and Mythos 5, for any foreign national. Because export rules treat distribution to any non citizen as an export event, the company could not apply the restriction selectively. The practical result was a global shutdown of both models for every customer, while access to its other models continued.

The trigger was a security finding. Researchers identified a technique that could bypass safeguards built into Fable 5, the publicly released model, and reach the deeper vulnerability finding capabilities of Mythos 5, the underlying system. That finding moved quickly to senior policymakers in Washington, who treated the cyber dimension as a national security matter. Anthropic stated that it viewed the issue as narrow rather than universal and expressed disagreement with a full recall, while complying with the order.

The reaction extended far beyond the companies involved. Political figures across Europe described the moment as a wake up call and renewed calls for sovereign AI, the principle that nations should control the advanced systems their economies increasingly depend upon.

Structural Context

This was not really a story about a single product. It was a demonstration that access to frontier compute can be governed, paused, or withdrawn by the jurisdiction that controls it. For European institutions that had begun building research, clinical, and commercial workflows on these systems, the disruption was immediate and concrete.

The deeper pattern is dependence. Advanced models are concentrated within a small number of providers operating under a single regulatory regime. When that regime acts, the effects travel worldwide within hours. Capability without control becomes a form of exposure, and the event made that exposure visible to policymakers who had previously treated AI as a commercial convenience rather than strategic infrastructure.

Europe holds genuine assets in this contest, including decarbonized power, strong engineering talent, and a growing roster of domestic technology firms. What the moment revealed is that owning ambition is not the same as owning capacity. Sovereign AI requires sovereign infrastructure, and infrastructure begins with the physical layer beneath the models.

The Enki Perspective

Sovereignty over models is the visible layer of a much larger question. Beneath every model sits compute, and beneath every unit of compute sits power. A nation cannot claim digital sovereignty while remaining dependent on infrastructure and energy it does not control. Digital sovereignty and energy sovereignty are the same objective viewed from two angles.

This is the structural opening that Project Enki is built to address. Across Europe, large volumes of renewable energy are produced and then curtailed, because transmission capacity cannot move that power to where demand sits. Curtailed energy is unrealized economic and digital capacity. Placing compute at the point of generation converts a stranded resource into sovereign capability, and it does so without waiting for grid expansion that takes years to clear.

The advantage compounds. Infrastructure that moves toward available power deploys faster, steps around interconnection delays, and presents institutional capital with a repeatable model it can finance at scale. Sovereign compute does not have to be assembled from imported dependence. It can be grown from energy that already exists within the region and is currently going to waste.

What This Signals

The lesson institutions are absorbing is that continuity now matters as much as capability. Procurement decisions once dominated by performance and cost are widening to include resilience, jurisdiction, and the question of who can interrupt a service. That shift favors architectures designed for regional control from the start.

Expect sovereign infrastructure to move from political language toward capital allocation. The constraint that defines the next phase of AI is not ambition or demand. It is access to power and the ability to deploy compute where that power is available. The regions that connect energy to compute most directly will set the terms of their own digital future.

Project Enki reads this moment as confirmation of a structural thesis. The future of advanced compute will be built where energy is abundant, and sovereignty will belong to those who control the full stack, from the power line upward.

Source: Robert Hart, June 2026 https://www.theverge.com/ai-artificial-intelligence/949986/anthropic-fable-mythos-shutdown-sovereign-ai