Blog · November 21, 2026
What to ask your vendor about data, AI, and Canadian compliance
Canadian buyers are now asking about data residency, storage and AI audit trails at procurement, not after. Here's how to frame the questions your vendors need to answer.

Canadian procurement has shifted. Twelve months ago, questions about where your vendor's AI tools stored data, how they processed it, and whether they kept an audit trail were asked late — sometimes after the contract was signed. Now they come early, often in the RFP itself or in the first screening call.
This is not an overreaction. It reflects two real changes. The EU AI Act's enforcement powers became active in August 2026, making audit logs and human oversight legal obligations for high-risk systems. And Canada's Treasury Board now treats data residency as mandatory in any AI risk assessment for federal departments. Quebec's Law 25 adds another layer: documented privacy impact assessments before personal data leaves the province. For procurement officers in government and large Canadian enterprises, the message is clear: know where your data goes before you sign.
If you are evaluating AI tools for an innovation pipeline, a business-case process, or any system where your team will input strategic or customer information, you need to ask four specific questions. Getting clear answers now — and getting them in writing — will save you months of rework and risk later.
Where does your vendor store the data?
Storage is the first control. When your team enters information into a vendor's system, where does it physically sit?
The right answer is usually "in Canada" — and more precisely, often "in Montréal" or "in a Canadian data centre" if you are serving Quebec or federal clients. A vendor headquartered in Canada does not automatically satisfy this. Many Canadian companies store data in US data centres for cost or scale reasons. That is not necessarily wrong, but it is a choice you need to see plainly.
Ask your vendor:
- Where is the primary data storage location? (Country and region, not vague.)
- Do you offer an option to store data in Canada, and at what cost or configuration?
- If data is replicated for redundancy, where do the replicas sit?
- Can you contractually commit that data never leaves a specific jurisdiction without consent?
Write down the answer. If your vendor says "it depends on your plan" or "we'll need to custom-build that," you now know you need a legal review and a specific contract clause. That is information worth having before you decide to move forward.
Where is the data processed when AI runs?
Storage and processing are not the same. Data can sit safely in Canada but travel elsewhere during inference — the moment when an AI model reads it and generates output.
This matters because processing is where privacy risks concentrate. When your business case or customer insight enters an AI system, the vendor's model can learn from it unless they contractually prohibit training on customer data. Some providers offer no-training guarantees; others do not. Some will add it to your contract for a higher fee or specific plan tier; others will not add it at all.
Ask your vendor:
- During inference — when your data runs through the AI model — where is the processing done?
- Do you have a contractual no-training clause that forbids the AI provider from using customer data to improve their models?
- If not, is that something you can add, and under what terms?
- Does your contract name the specific AI providers (Anthropic, OpenAI, Google, xAI) whose services process your data?
- Are there any scenarios — debugging, model improvement, compliance checks — where your data could be accessed outside the jurisdiction where it was stored?
Again, write down the answer. If your vendor cannot tell you which AI provider processes their inference, or if they cannot contractually commit to no training on your data, that is not automatically disqualifying — but it is a material risk you need to disclose to your legal team and your approvers.
Does your vendor keep an audit log of every AI call?
Governance is now enforceable, not optional. The EU AI Act requires audit trails for high-risk AI systems. Canada's Treasury Board guidance expects the same. And if your organization works with federal departments or Quebec-regulated clients, they will ask for proof that every AI decision can be traced.
An audit log answers one question: what happened, and who approved it? For each time an AI tool generated output — a draft answer, a research summary, a document section — the log should record:
- Which AI model ran (Claude, ChatGPT, Gemini, Grok).
- What prompt was used (the instructions the vendor gave the AI).
- What input the model received (your data).
- What output it produced.
- Who accepted, edited, or rejected it.
- When this all happened.
This is not paranoia. It is due diligence. If a client later asks "how did you arrive at that conclusion?" you need to be able to show the work — and show that a human decided to accept it.
Ask your vendor:
- Do you keep a full, append-only audit log of every AI call?
- Can you show me a sample of what that log contains?
- How long do you retain these logs, and can I access them on demand?
- If there is a compliance review or audit, can you export the full log for a specific idea or project?
- Is the log stored encrypted, and who has access to it?
If your vendor hedges ("we log some calls," "logs are available on request but we charge for retrieval," "we keep logs for 90 days"), that is a red flag. Logs should be comprehensive, free to access, and retained long enough to satisfy your own record-keeping requirements — usually 7 years for regulated industries.
What about confidential information?
If your organization has ideas, business cases, or customer data that cannot touch any external AI service under any circumstances, ask this directly:
- Do you ever send confidential information to external AI providers, even with safeguards?
- Can you contractually commit that confidential ideas will never be processed by external AI, with no exceptions and no override?
- If a user tries to run AI on a confidential idea, does the system block it or warn them first?
Some vendors allow confidential information to go to AI providers but claim they have contractual protections. Others keep confidential work off external AI altogether — the information stays on their own servers and runs on their own infrastructure. That second approach is stronger, and if your work is truly sensitive, it is worth looking for vendors who offer it.
Putting it together
Before you sign a contract or add a vendor to your procurement list, build a simple matrix: storage location, processing location, training prohibition, audit-log completeness, and confidential-data handling. Ask each vendor to fill it in. Have your legal team review the answers.
This is not about finding a perfect vendor. It is about knowing what trade-offs you are making. A vendor with data stored in the US but comprehensive audit logs and a strong no-training clause might be acceptable to your compliance team. A vendor with Montréal storage but weak audit logging might not be. The choice is yours — but it should be an informed one, made at procurement, not discovered in a compliance review six months in.
The ground has shifted. Data residency and AI governance are no longer nice-to-have features. They are table stakes for any vendor working with Canadian organizations. Asking these questions now signals to your vendors that you take this seriously, and it gives you the information you need to make a decision your board and your clients will trust.