Supplier questions

The eight control decisions, put as questions to a supplier

An AI architecture review that asks only where the data sits has answered one question in eight. This is the other seven questions, phrased as you would ask them, with the answer that should prompt a follow-up. Take it into the meeting.

Take one workflow through the eight controls Read the sovereignty design whitepaper

The short answer

Is Canadian data residency the same as AI sovereignty?

By Dan Dorion, Chief Executive Officer · Published 15 August 2026

No. Data residency identifies where data is stored or processed. AI sovereignty is broader. It includes legal jurisdiction and the ownership and control of infrastructure. Administrator and support access count too, as do encryption keys. It also covers model provenance, telemetry and updates, plus incident response and the ability to continue operating. A Canadian region operated by a foreign-controlled provider may meet a residency requirement without meeting every sovereignty objective. The right design depends on the workload and its data classification, the procurement terms and the risk assessment.

This page is the supplier-facing half of that answer: for each of the eight controls, the question to ask, and the answer that should prompt a second question. None of it concludes that any particular architecture is wrong, only that one question out of eight is not enough to describe what an institution has bought.

Note

Which of the two to read

If you are evaluating a supplier or sitting in a vendor meeting, this page. If you are designing or costing an architecture, read the sovereignty design whitepaper. It defines each of the eight controls in full and carries the architecture options, the shared-responsibility split and the cost assumptions. The eight are the same eight, defined there and applied here.

Use it in a supplier conversation

The question to ask, and the answer that should worry you

Take this table into a supplier meeting. Find out quickly which of the eight controls the supplier documents and which it assumes. The sovereignty design whitepaper defines each control in full.

Eight control decisions with a specific supplier question for each and the answer pattern that should prompt a follow-up. Generic guidance; no supplier or product is characterized as compliant or non-compliant.
Control	The question to ask	An answer that should worry you
01 Data location	For this workload, where do prompts, retrieved passages, embeddings, outputs, logs, caches and backups physically reside? Can you show me that in documentation?	“Everything stays in Canada” with no distinction between the primary store and derived artifacts, and no document behind it.
02 Legal jurisdiction	Which legal entity are we contracting with, where is it incorporated, and which entities in your group could be subject to legal process relating to our data?	An absolute assurance that no foreign law could ever apply. That is a legal conclusion no supplier can responsibly give you across all circumstances.
03 Infrastructure control	Who owns and operates the infrastructure, and can you list the subprocessors and hosting locations involved in this service?	An unwillingness to name subprocessors, or a list that has no change-notification commitment attached.
04 Identity and support access	Which roles can access our data or environment, from which countries, under what approval, and how is that access logged and made available to us?	“Nobody can access your data.” Someone always can, under some break-glass procedure. The useful answer describes the procedure and the record.
05 Encryption and keys	Who holds the keys, who can technically use them, how do we rotate or revoke, and what precisely happens to our data if we do?	Customer-managed keys described in a way that turns out to mean the provider retains an operational path to plaintext.
06 Model and software	Can we pin a model version, how much notice do we get before it changes, and is our content used for training or product improvement?	No version pinning, or a training answer that lives in a policy page the supplier can revise unilaterally rather than in the contract.
07 Telemetry	What does the service emit about our usage and content, to whom, for how long, and which of it can we reduce or turn off?	A confident “we retain nothing” with no reference to safety systems, abuse monitoring or operational logging.
08 Continuity and exit	In an incident, what evidence can we obtain and when? If we leave, what comes back, in what format, and how long do you retain it afterwards?	Exit described as an account-closure process rather than a data-return process with formats, timelines and deletion attestation.

What if a supplier cannot answer all eight?

Treat that as information about the state of the service. Some controls are unsettled for new services, and a supplier who says so is more useful than one who improvises. Record the gap, decide whether it is material for this workload at this classification, and either accept it explicitly with a named approver or design around it. What you should not do is let an unanswered question pass into production as an assumption nobody wrote down.

Judgement

Proportionality, and where residency is enough

This framework exists to make the weight land where the risk is, not to make every project heavier.

For an internal assistant over already-public information, residency plus a clear position on training and telemetry may be the whole of the analysis. Demanding key custody arrangements for public information adds months without reducing risk. For a workflow touching case files, personal information or material subject to a statutory confidentiality obligation, the other seven controls do the work. The residency answer on its own tells you little.

The decision that connects the two is classification and decision impact, which is why we help the institution screen for both before any architecture discussion. That sequence (classify, screen, then choose the pattern) is set out in the Canadian Sovereign Agent Deployment Playbook. The applicability screen for the Directive on Automated Decision-Making is in the Algorithmic Impact Assessment (AIA) guide.

Note

Two habits that survive any architecture

Write the answers down with the date you received them and the document they came from. Supplier positions change, and an undated assurance ages into a liability.
Re-ask on a schedule and on change. A new model, a new connector, a new region or a new subprocessor can move three of the eight without anyone on your side being told.

Honesty

Limitations and what this does not cover

This is not legal advice. Jurisdictional conclusions depend on the entities, contracts and circumstances in your specific arrangement and require your own counsel. We make no claim about how any foreign legal process would apply to any provider, and you should be sceptical of anyone who does.
No supplier is assessed here. We name no provider as compliant, non-compliant, sovereign or otherwise. The framework is for evaluating suppliers; it is not an evaluation of any.
We make no absolute claims. Not about retention, not about exposure to any legal regime, not about any architecture being immune to any risk. Where you see such a claim in this market, ask for the document behind it.
The eight controls are a framework, not a standard. They are our own structuring of the questions and carry no external accreditation. They are designed to sit alongside your institution’s security assessment, privacy assessment and procurement process, not to replace any of them.
Nothing here is a certification claim about White Glove Labs. What we can and cannot evidence today is set out in the supplier evidence pack, sent on request.
Accurate to our reading on 15 August 2026. Provider terms and Canadian policy both move; verify before relying on this in a submission.

How we know this

Primary sources

Source

The eight-control framework is our own. The Canadian instruments it is designed to work alongside:

Treasury Board Directive on Automated Decision-Making, modified 24 June 2025.
Office of the Superintendent of Financial Institutions (OSFI) Guideline E-23, Model Risk Management, effective 1 May 2027, covering AI and machine-learning models including third-party models.
Innovation, Science and Economic Development Canada (ISED), AI for All national strategy, June 2026, organized around trust, opportunity, sovereignty, adoption, Canadian infrastructure, plus scaling Canadian champions.
Policy on Prioritizing Canadian Suppliers and Canadian Content in Strategic Federal Procurements, applying to strategic federal procurements at CAD $5M and above from 15 June 2026.
Public Services and Procurement Canada (PSPC) Artificial Intelligence Source List.

Next step

Run the eight questions against one real workload

The exercise takes an afternoon. Any cell nobody can fill in belongs on the risk register.

Source

Dollar figures on this page are the sources’ own or recorded amounts, not our prices. Our prices are in Canadian dollars (CAD), plus GST/HST.

Bring the workflow, the classification if it is known, the suppliers already in the picture, and the officer who will have to review it. We will work the table with you and leave you the completed version, whether or not we go further together.

Book a 30-minute call See the full sovereignty architecture Request the supplier evidence pack