A connector needs a measure, not an adapter

On 3 August the Financial Times published a column by Roland Busch, chief executive of Siemens, under the headline “A digital iron curtain is threatening the global economy”. Its standfirst names a role Europe could take: a trusted connector between technology ecosystems that are drifting apart.

The diagnosis is right. For three decades digital technology stitched the world together, and almost no complex product is now made in a single place. Today globalisation is becoming more regional, and digital infrastructure is turning into an instrument of pressure. The connector role is a sensible answer to that.

But the formula is missing one component, and without it the role cannot be performed.

Interoperability settles syntax, not trust

Interoperability answers whether two systems can exchange data. It does not answer whether they should.

Where two ecosystems do not trust each other, a shared exchange format does not create trust. It creates a channel. A connector without a measure of trust is not a bridge — it is an open door between two perimeters that regard each other as hostile.

This is not a rebuttal of the connector thesis. It is the condition under which it works. A connector needs a measure, not compatibility.

Why a sovereign stack is not the answer

The temptation is obvious: if we depend on other people's solutions, let us build our own. Tellingly, the same Siemens chief executive has previously called it a disaster to prioritise sovereignty at the cost of innovation speed. He is right. A European copy of somebody else's stack does not make Europe a connector — it makes Europe a third bloc.

A measure of trust is different in kind. It does not replace models and platforms; it sits above them. It does not ask any ecosystem to migrate to another's stack. It asks only one thing: that claims about behaviour be verifiable by someone with no stake in the answer.

That is the difference between building your own and making other people's things work together.

Neutrality requires a non-owner

Neutral measurement is impossible for whoever owns the thing being measured.

A platform cannot be a neutral arbiter of trust in the agents running on it. A payment network cannot be a neutral arbiter of counterparty risk in the transactions it settles. This is the same structural conflict that placed rating agencies outside banks rather than inside them.

From which follows a non-obvious conclusion about Europe. Its advantage here is neither regulatory will nor market size. Its advantage is that it has no hyperscaler whose trustworthiness would have to be measured. What is usually described as a weakness is, in the role of arbiter, the only qualification that matters.

A measure that can refuse

The property that makes a connector trusted rather than merely permeable is the willingness to say no.

A measure that produces reassuring numbers from thin evidence is worse than no measure at all. It manufactures false confidence where there was honest uncertainty. The primary requirement for such a system is therefore not the accuracy of its scores but its capacity to decline to score when the evidence is insufficient.

It is an inconvenient property. It makes the product less impressive. It is also the only reason such a measure could be trusted at all.

Who or what: a line English lets you miss

Note the sentence three paragraphs above: “whoever owns the thing being measured”. In English that reads naturally. In Ukrainian, my first language, it cannot be written without a decision — Slavic grammar obliges the writer to mark the measured party as animate or inanimate, as a who or a what. The question English permits you to skip is one my language forces.

And once forced, the usual answer turns out to assert something. Calling an autonomous agent a “what” settles in advance that it carries no history of its own — which is precisely what a trust measurement exists to read. You can only measure a party that has a record.

I am not proposing legal personhood for agents. The “electronic person” idea was floated in the European Parliament in 2017 and abandoned after an open letter from over a hundred European researchers. That rejection was correct. And there is a serious objection to moving the other way: work published this year by Petersen and Almor finds that placing an AI in the agentive grammatical position leads people to assign more responsibility to the AI and less to the company that built it. The grammar is not decoration; it redistributes blame.

So the line has to be drawn precisely. An agent is a subject of measurement — the bearer of a behavioural record that can be appealed to. It is not a subject of law: responsibility stays with whoever deployed it. These are two different claims, and conflating them is dangerous in both directions.

But calling it a thing asserts that there is nothing there to measure. On that premise the whole construction collapses before the first number.

What follows

Europe can indeed become a trusted connector. But “trusted” in that phrase is not an adjective — it is a technical requirement. It means a measure that belongs to none of the parties and that knows how to refuse.

Building such a measure is harder than agreeing on an exchange format. It is also the only way the role named in that column stops being rhetoric.