Institutional readiness is the real AI policy question

The most interesting AI policy question is not which rule to adopt. It is whether the institution being asked to adopt it has the internal capacity to read, commission, and contest the systems it is meant to regulate.

Capacity before compliance

When a ministry sets out to regulate a capability it does not yet understand, the result is not usually failure. It is something slower and more corrosive: a regulatory apparatus that looks present on paper and is absent in practice. The regulated entities operate. The rules do not bind. Public reasoning thins.

This is the pattern to avoid. And it is the pattern most likely, because compliance is easier to legislate than capacity.

What AI-competent institutions actually look like

The institutions that will do this well in the next decade will share three characteristics. They will have internal technical literacy at the senior level, not only in a research unit. They will have procurement regimes that can evaluate systems under real operating conditions. And they will have an editorial culture, inside the institution itself, that can write about what it does in public-interest language.

An AI-competent ministry is not one that has adopted an AI strategy. It is one whose senior officials can read a system the way a finance ministry reads a budget.

The next two years

The next twenty-four months will separate institutions that treated AI as a communications question from those that treated it as a capacity question. The first group will produce strategies. The second group will produce results that survive contact with the systems they deploy.

Neither is a criticism of any particular regulator. It is an observation about the kind of work that has to happen inside the building, before policy even leaves it.