The Copilot problem nobody talks about
Every client who asks us to deploy Microsoft 365 Copilot gets the same first question from us: “What’s your SharePoint permission model look like?”
The answer is almost always some variation of “well, it’s fine, we think.”
It’s not fine. And that’s the Copilot problem nobody talks about.
Copilot is an AI layer that surfaces content from across your Microsoft 365 environment — SharePoint, Teams, Exchange, OneDrive, Loop. It respects your existing permission model. Which means if your permission model is “everyone can read everything because it was easier to set up that way,” Copilot will cheerfully summarize a meeting transcript from a confidential board discussion and surface it in response to a query from an intern.
This isn’t a bug. It’s the expected behavior. Copilot sees what you’re allowed to see.
The four governance moves
Before we turn Copilot on for any client, we run four governance steps — in this order.
Step 1: Oversharing audit. We run a SharePoint permission audit using the Microsoft 365 data governance toolkit and our ISMS platform. This shows us every site, library, and document that has permissions broader than the team that owns it. The typical SMB has somewhere between 40 and 200 overshared objects. We remediate before Copilot is enabled.
Step 2: Sensitivity labeling. We configure Microsoft Purview sensitivity labels that map to your data classification policy — Confidential, Internal, Public. We apply auto-labeling rules for documents containing PII, financial data, or legal content. Labels restrict Copilot from surfacing labeled content to users who don’t have access to the underlying label.
Step 3: Copilot plugin governance. Copilot can be extended with plugins that connect to external systems — Salesforce, Jira, HubSpot. Each plugin is a potential data path you didn’t design. We configure the Copilot admin center to require IT approval for any plugin that connects to external data sources.
Step 4: User training and prompting guidelines. Copilot’s output is only as good as the prompt and only as trustworthy as the user who reads it. We run a 45-minute session for each client cohort covering what Copilot can and can’t access, how to write effective prompts, and the three things to verify before acting on Copilot output (source, recency, context).
The insurance question
When a cyber insurer sees “Microsoft 365 Copilot” in a security questionnaire, they want to know two things: (1) is sensitive data accessible through it, and (2) do you have a policy governing its use?
The answer to (1) is determined by your permission model and labeling. The answer to (2) is your Copilot acceptable use policy — a one-page document we author as part of the TruCompliance ISMS engagement.
Both need to be in place before renewal, not after.
The Canadian data residency piece
Copilot processes queries on Microsoft’s infrastructure. Microsoft has committed that Copilot respects data residency commitments for Microsoft 365 — if your tenant is provisioned in Canada, Copilot query processing stays in Canada.
Verify this is true for your tenant by checking the Microsoft 365 admin center under Settings > Org settings > Organization profile > Data location. It should say “Canada” for all workloads.
If it doesn’t, you need a conversation with your Microsoft account team before Copilot goes live.
Alex Tremblay is the CTO at TruPoint Technology.