We Just Saw the AI Kill-Switch in Action — And Why It Makes EU Sovereignty Non-Negotiable

A few days ago, something a lot of us in tech had quietly filed under "hypothetical" became real.

What happened

On June 12th, 2026, the US government forced Anthropic to take two of its AI models — Fable 5 and Mythos 5 — offline. Not throttled, not geo-blocked in one region: shut down for customers worldwide, through a single export-control directive. One decision, in one jurisdiction, and a tool that businesses and products around the world were building on was simply gone.

The official reason — and why it is contested

The directive cited national security. The authorities claimed there was a way to "jailbreak" Fable 5 and bypass its safety mechanisms — but published no technical detail to back it up.

Anthropic pushed back hard, calling the move a misunderstanding. It demonstrated that the alleged jailbreak — asking the model to read code and fix errors — surfaced only minor vulnerabilities that publicly available models like GPT-5.5 reveal too. The shutdown also did not come out of nowhere: in March 2026 the US Department of Defense had already classified Anthropic as a "supply chain risk", a label CEO Dario Amodei calls "legally untenable" and intends to challenge in court.

You can debate the merits all day. That is not really the point.

This is a kill-switch — and we just watched it work

Strip away the specifics and one fact remains: a service that organisations across every continent depended on was switched off by a decision made in a single country.

For years, the idea of an AI "kill-switch" was waved away as paranoia or science fiction. We have now seen one used in the real world. Nobody can credibly deny it anymore.

The uncomfortable question for anyone running a business on someone else's infrastructure: it was not your model, your contract, or your uptime SLA that decided this. It was a government you do not vote for, applying laws you are not party to.

Why this is a European problem

If the layer your business runs on can be turned off by a regulator on another continent, then "it works today" is not a strategy — it is a risk you simply have not been billed for yet.

This is the same structural dependency we have written about before: most European companies build their entire backend — auth, database, storage, payments — on US-controlled infrastructure subject to US law, including the CLOUD Act. An export-control shutdown of a model is just the most visible version of a much broader exposure.

The EU is on a different path

It is worth noting what did not happen. A comparable single-model shutdown is not really on the table inside the EU. The EU AI Act takes a risk-based approach — obligations scaled to how a system is actually used — rather than an export-control on/off switch wielded for geopolitical leverage. Different philosophy, different failure modes, and for European builders, a meaningfully different risk profile.

What we are doing about it

We will not pretend one startup solves sovereignty on its own. But we believe every part of the stack that stays in Europe is one less switch someone else gets to flip.

That is why we built Eurobase: a fully EU-native Backend-as-a-Service (BaaS) platform. Auth, Postgres, storage, realtime, and serverless functions — all on European infrastructure, EU-owned and EU-hosted, GDPR by design, and outside the reach of the CLOUD Act. Think Firebase or Supabase, without the American jurisdiction attached.

Build on foundations you actually control.

What happens to your product if your provider goes dark tomorrow? Sign up free.