Ten sovereignty questions to put to a security vendor

These three place a vendor in a category before any product discussion, and they are the ones most likely to be answered vaguely if asked verbally.

Who is the contracting entity, and who is its ultimate parent? Has the ownership changed in the last three years, and are you notified if it changes during the term? Which entity holds the encryption keys, and can any other entity in the group direct that entity to produce them?

The third is the one that separates a considered answer from a rehearsed one. A vendor who has thought about sovereignty will have an answer about key custody; one who has thought about marketing will return to the hosting region.

Three more, and these are where residency statements are usually complete on storage and silent on everything else.

Which regions handle storage, processing and backup, and does failover ever land outside them? Which countries do support and administration staff operate from, and what can they read in a production incident? What product telemetry leaves the region, and is it content or metadata?

The support question is the one with the largest gap between the written answer and the operational reality. Ask specifically whether an engineer outside the EU can access production data during an escalation, since the answer is often yes with an approval step that nobody counts as a transfer.

The FTC has accumulated vast rulemaking, enforcement and adjudicatory powers, and it unquestionably exercises executive power, and must therefore be controlled by the Chief Executive, in whom such power is vested.

Where does inference execute, under whose contract, and are you told if the provider changes? Are prompts and outputs retained, by whom and for how long?

They are separate because a vendor can have a good answer to the first and never have asked the second. Inference in an EU region with thirty-day prompt retention at the provider is a different exposure from the same region with zero retention, and the difference is contractual rather than technical.

If bring-your-own-model is offered, the questions move to your own procurement, which is usually an improvement for sovereignty and additional work for you. Worth knowing which of the two you are choosing.

What does an export of our data look like, in what format, and how long does it take? What happens to our data if we leave, and how is deletion evidenced?

These are the sovereignty questions that survive contact with reality, because they determine whether any of the earlier answers can be acted on. An arrangement you cannot leave is one where the ownership answer stops mattering the moment it changes for the worse.

For financial entities this is not optional. DORA Article 28 requires exit plans for arrangements supporting critical or important functions, and the same discipline is worth borrowing whether or not you are in scope.

File them with a date. Ownership, subprocessors and inference arrangements all change, and an answer from eighteen months ago describes a company that may not exist in the same form.

Score them against each other rather than against an imagined ideal vendor. The one who says plainly that inference runs outside the EU and explains the retention terms is more useful than the one who says sovereign and declines the detail.

Then ask the question that is not on the list: what would make you tell us we are not a good fit? A vendor with a real answer names a data classification, a sector or a scale where their arrangement is the wrong one. A vendor without one has described a product that suits everybody, which no product does.