Microsoft 365 Copilot is now showing up inside the firm tenants of Ontario law practices without a formal rollout decision ever being made. It ships bundled into many Microsoft 365 E3/E5 add-ons, lawyers turn it on in Word and Outlook because the button is already there, and nobody asks whether client file data just became the subject of a new AI processing activity. It did. Under PIPEDA, any client information a lawyer feeds into Copilot, and anything Copilot generates about an identifiable person, is personal information the firm is responsible for, the moment the prompt is typed, not at some later review stage.

The short answer to the headline question is: yes, Microsoft 365 Copilot can be used in a way that is consistent with PIPEDA and the Law Society of Ontario’s confidentiality rules, but only if the tenant is actually configured for it. Out of the box, most firms are running Copilot on default settings that were never built with a regulated professional’s confidentiality duty in mind. Below are the six settings that make the difference, and why each one matters.

What PIPEDA Actually Requires Once Copilot Is in the Tenant

PIPEDA does not have a Copilot-specific clause, but it does not need one. The statute requires organizations to obtain meaningful consent for the collection, use, and disclosure of personal information, to limit collection to what a reasonable person would consider appropriate, and to be able to explain how that information is used. Copilot’s grounding process, where it searches a user’s accessible mailbox, SharePoint sites, and Teams chats to answer a prompt, is a collection and use of personal information every time it runs, even when the output never leaves the tenant.

On top of PIPEDA, the Office of the Privacy Commissioner of Canada, together with every provincial and territorial privacy regulator, published joint “Principles for responsible, trustworthy and privacy-protective generative AI technologies” on December 7, 2023. The principles are not binding law, but they are the lens regulators will use when something goes wrong, and they say plainly that an output generated about an identifiable individual, including a Copilot summary of a client’s file, counts as a new collection of personal information requiring its own legal authority. The firm, not Microsoft, carries that accountability.

Setting 1: Advanced Data Residency and the Preferred Data Location

Microsoft’s standard Data Residency commitment covers core workloads, Exchange, SharePoint, OneDrive, and Teams, for tenants provisioned in the Canada geography. Copilot interaction content, meaning the prompt and the response, is covered separately and only when the tenant has an Advanced Data Residency (ADR) subscription and the Global Admin has actively opted in to move existing data into the Canada region. A firm that signed up for Microsoft 365 years ago, before ADR existed, is not automatically covered just because its billing address is in Ontario.

Storage location and processing location are also two different Microsoft commitments that get conflated constantly. Microsoft’s own documentation uses the example of a user in Canada asking Copilot to rewrite a paragraph in a document that physically lives in France: the prompt and the rewrite are stored in Canada, but where the underlying inference computation happens is a separate question. Microsoft announced in November 2025 that in-country data processing for Copilot would extend to Canada during 2026, with full local inferencing following in 2027. Until that lands, a Canadian tenant with ADR enabled is buying Canadian data storage with inference that may still run in a US or EU data centre.

Setting 2: Purview Sensitivity Labels on Matter and Client Folders

Copilot only respects confidentiality boundaries it can see. Microsoft Purview sensitivity labels, applied manually or through auto-labeling policies, are what tell Copilot a file or folder is restricted before it ever becomes part of a grounding search. A firm running without a labeling taxonomy is relying on Copilot to apply the same judgment a lawyer would about what belongs in a specific matter, which it cannot do. Auto-labeling for sensitive content types requires an additional Purview licence tier, which is the step most firms skip after buying Copilot seats and stop there.

Setting 3 and 4: DLP for Copilot, Grounding and Prompts

Purview’s Data Loss Prevention policies for Copilot operate at two separate points, and both need to be configured, not one or the other.

⚠ A firm that enables one of these two and assumes it covers the other has a gap. Grounding DLP protects what Copilot can read; prompt DLP protects what a user can type. A junior associate pasting a client’s financial statement directly into a Copilot chat window bypasses grounding DLP entirely, because they supplied the data themselves.

Setting 5: Audit Log Retention and Data Lifecycle Management for Copilot Interactions

Every Copilot prompt and response generates an auditable record. The question is how long the firm chooses to keep it, and whether it can produce that record on request. PIPEDA’s accountability principle and the OPC’s generative AI principles both expect an organization to be able to explain and evidence how an AI tool was used in connection with a specific matter, not just assert that it was used responsibly. Microsoft Purview’s eDiscovery tools can search, preserve, and produce Copilot-related content for an audit or legal request, but only if a retention policy was set before the interaction happened. By default, a firm has no documented answer to “what did Copilot see on this file, and when.”

Setting 6: Compliance Manager’s PIPEDA Assessment

Microsoft Purview Compliance Manager includes a pre-built PIPEDA assessment template that scores a tenant’s actual configuration against the statute’s requirements and against Microsoft’s own recommended actions, rather than against generic best practice. Most firms never open it. Running it once produces a documented, dated baseline a managing partner can point to if a client, insurer, or the Law Society ever asks how the firm evaluated Copilot before deploying it, which is the accountability evidence the OPC’s principles specifically call for.

Where This Fits Next to LSO Confidentiality Obligations

None of the above replaces a lawyer’s own supervisory duty. Rule 3.3-1 confidentiality and Rule 3.1-2 competence still sit above the technical configuration, and the Law Society’s expectation is that a lawyer using any AI tool, Copilot included, verifies the output and can explain how client information was handled. What the six settings above do is make the technical environment match that standard, instead of leaving a lawyer’s professional judgment as the only control in a system that was configured on Microsoft’s defaults. A firm that has done the Purview and Compliance Manager work has something concrete to show a client, an insurer during a cyber insurance renewal, or the Law Society during a practice review. A firm running on defaults has only an assurance that nothing has gone wrong yet.

Frequently Asked Questions