Change / See / Stop

Frameworks / Change / See / Stop


Residency is an address. Sovereignty is a veto. Three questions, asked of each AI workload, tell you which one you actually hold.

Most sovereignty conversations end at one checkmark: the data stays in the country. That answers where something is stored. It does not answer who can reach it, who can change what it does, or who can switch it off. As AI pulls an organisation’s process into prompts, retrieval, fine-tunes and agent workflows, those are the questions that decide whether the intelligence is yours. Data is the raw material. Process is the recipe, and locality does not protect the recipe.

The three questions

Change: who can alter it without asking you? Model updates, deprecations and policy changes alter how your process behaves. Whoever controls that schedule controls your recipe.

See: who can access it, and under whose law? Prompts, retrieved context, outputs and logs are where the recipe becomes visible. Where they are stored matters less than who can compel access to them.

Stop: who can switch it off, and what still runs if they do? A provider, a regulator or another government can end access on its own timetable. Continuity is the question a residency clause never asks.

WHERE residency: the address, answered once answered only us our provider CHANGE SEE STOP updates · deprecations · policy prompts · logs · whose law switch-off · what still runs sovereign rented answered per workload, not per platform

A workload that answers “only us” to all three is sovereign. One that answers “our provider” to all three is rented, and that may be exactly the right choice. Most workloads land somewhere in between, and the mix is the point: it shows which of the three you would need to fix if the workload ever had to become yours.

How to see it in the field

The tell is a sovereignty answer that is a location. Ask who controls a critical AI workload and hear “the data is in-country”: the address has been mistaken for the veto. A second tell is a model update that arrives as a surprise, discovered when a process starts behaving differently rather than announced to the people who own that process. A third is a continuity plan that covers a data centre failure but not a provider, or a jurisdiction, withdrawing a model.

Using it well

Ask the three questions of each workload, not of the platform as a whole. The Workload Sovereignty Matrix tells you which workloads should be sovereign; this test tells you whether they are. The Intelligence Control Stack shows which layer is rented when an answer comes back as someone else.

The test does not grade the answer. Renting is often correct, and bringing everything home is the same unexamined decision made in the opposite direction. It applies to every option, including a private stack on your own floor: closed weights updated on a vendor’s schedule, or support delivered from another jurisdiction, fail it there too. What it guarantees is that someone gave an answer, on purpose, and owns it. In the Exit Multiple, these three answers are the Sovereignty ledger.


Part of the Frameworks series. First argued in Residency is an address. Sovereignty is a veto. Related: the Workload Sovereignty Matrix, the Intelligence Control Stack and the Context Layer.