On March 17, Sweden's Foreign Minister flew to Houston and signed Pax Silica. No prior parliamentary decision. No public debate. No explanation of what Sweden had actually committed to. The EU had not taken a position. The Foreign Minister did not answer questions from the press, she sent a text message through her press secretary.
Eleven Swedish tech figures sent an open letter with five concrete questions. None of them have been answered.
The debate that followed was about geopolitics: EU versus US, Ericsson loyalty, democratic deficit. All of that is right. But it's the wrong question.
The right question is architectural.
Sovereignty Is Not a Flag, It's an Inference Chain
In an earlier article, I argued that Sweden is building an administration where algorithmic opacity functions as legal shield. That the fastest path to AI in the state is also the path that tears down the accountability infrastructure it took generations to build.
Pax Silica is the next chapter in the same story.
The declaration covers semiconductors, critical minerals, energy, connectivity, and by extension AI models and digital infrastructure. It is not a trade agreement. It is an agreement about who controls the technical stack, and the closed AI platforms, on which future decisions will be made.
And here is the problem no one in the debate has put words to: you cannot have traceable, auditable decision-making in public administration if the infrastructure those decisions are made on is controlled by the strategic interests of a foreign state.
That's not a political point. It's an architectural reality.
The Kiruna Question Is Really an Administrative Question
Critics ask about LKAB's deposits in Kiruna. That's a legitimate question. But it's still the wrong layer.
Think about it this way instead: Sweden wants a hundred government agencies to use AI in their decisions. Sweden wants those decisions to be auditable, appealable, and justifiable. Sweden wants, according to its own AI strategy, to be best in the world at using AI in public administration.
At the same time, Sweden is signing a declaration on cooperation over digital infrastructure. The declaration is public. The questions concern its implementation: which actors get access to what, and how will trade-offs between national and American interests be handled when they collide?
You cannot demand of the Swedish Public Employment Service that it be able to explain why an algorithm rejected a job seeker, if the algorithm runs as a closed service on a platform whose terms, logs, and model updates are controlled by a foreign actor.
Accountability ends where control ends.
The Worst Part Isn't That Sweden Signed, It's That No One Understood the Connection
Erik Slottner spoke about digital sovereignty in Berlin last autumn. Three months later, the same government signed Pax Silica without debate. That's not hypocrisy. It's something worse: it's a sign that the two questions aren't even perceived as related.
Digital sovereignty is discussed as a geopolitical question, about independence from the US, about EU loyalty, about Ericsson's market access. Algorithmic accountability is discussed as an administrative question, about transparency, about appealability, about human control.
They are the same question.
If you don't control the stack, you can't control the inference. If you don't control the inference, you can't enforce accountability. If you can't enforce accountability, you don't have a state governed by the rule of law, you have one that looks like one.
The Question No One Is Asking
The open letter to the Foreign Minister contained the right questions about process and obligations. But it was missing the question that is most central for every citizen whose case is one day decided by an algorithm:
How does Sweden ensure that the AI systems used in the exercise of public authority can be audited, reviewed, and held accountable as the commitments in the public Pax Silica declaration are put into practice?
That is not a technical question. It is a constitutional question.
And until it is answered, every administrative decision made with the support of AI on closed foreign platforms should be treated as lacking legal certainty, not as a technical problem to be solved later.
Best in the world at AI in public administration. That is still the ambition.
But you cannot build a responsible AI society on a stack you don't own, through a process you didn't understand, via an agreement you didn't debate.
Sovereignty is not a flag you raise. It is an inference chain you control.
: The Pax Silica declaration is public. Statements about its availability have been corrected.