When the AI kill switch becomes geopolitical.

The real story is not whether one model was safe. It is who can switch it off.

The Anthropic episode is being reported as an export-control intervention, an AI-safety dispute, a corporate crisis and a European sovereignty alarm. Each frame is partly right, but none is sufficient on its own.

For Washington, the action signals that frontier AI is no longer only a product category. It is strategic infrastructure, treated closer to chips, encryption or dual-use technology than to ordinary software. For Anthropic, the awkward lesson is that arguing for strong AI governance also invites a state to decide when governance becomes control. For Europe, the lesson is sharper still: dependency is not theoretical when access can be removed by a letter received late on a Friday afternoon.

The most considered reading is therefore not that America has become unreliable, nor that Europe can simply regulate its way to sovereignty. It is that AI power now sits at the intersection of model safety, cloud infrastructure, export law, energy capacity and diplomatic trust. The policy question is no longer whether Europe wants sovereign AI. It is whether it is willing to build the dull, expensive and necessary foundations that make sovereignty more than a slogan.