No single Canadian body regulates “AI in the professions.” Courts, law societies and privacy regulators each pull a different lever, and a professional needs to know which one is actually engaged by what they are doing.
Key takeaways
Ask “what has Canada said about AI in the professions” and the honest answer is: no single body has, because there is no single regulator with that mandate. What exists instead is a stack of separate regulators — courts, law societies, privacy commissioners, and at least one health-information regulator — each speaking from its own lever, on its own timetable, to its own audience.
The Federal Court, Alberta’s three courts, and Ontario’s Superior Court of Justice have each issued their own AI notice or practice direction, and each governs only its own proceedings. The Federal Court requires a Declaration where AI content “resembles that of a co-author”; (Federal Court, Notice to the Parties and the Profession, May 7 2024) Alberta’s tri-court notice asks for caution and verification with no disclosure duty; (Alberta Courts, Tri-Court Notice) Ontario’s Consolidated Practice Direction adds a formal sanctions regime for “hallucinations” that reach a filing. Three courts, three distinct texts, covering the same underlying worry from three different angles — none of them binds a proceeding in a different court.
Separately, nine Canadian law societies (by the Law Society of Alberta’s own count) have issued professional-conduct guidance covering confidentiality, competence and billing — obligations that apply whether or not AI content ever reaches a court. (LSA, Gen AI Rules of Engagement) This is a genuinely different lever from a court notice: a lawyer can fully satisfy a court’s verification expectation on a specific filing and still breach a law society’s confidentiality guidance by pasting client facts into a public tool while drafting it.
A third lever runs underneath both of the others and applies outside the legal profession entirely. Canada’s federal, provincial and territorial privacy commissioners jointly published generative-AI principles in 2025 covering fairness, necessity, accuracy and accountability for any organization deploying the technology — not lawyers specifically. (OPC, generative AI principles, 2025-05-06) Alberta’s OIPC has gone a step further for one profession by name: its AI Scribe Privacy Impact Assessment Guidance, tied to section 64 of the Health Information Act, is written specifically for healthcare custodians and reads nothing like the courts’ or the law societies’ guidance. (OIPC AB, AI Scribe PIA Guidance)
For a licensed professional handling a client or patient matter with AI assistance, the question is never “is this allowed” in the abstract — it is which of these layers actually applies to what they are doing. Confidentiality obligations (privacy regulator and professional body) apply regardless of whether a court is ever involved. A court’s disclosure or verification rule attaches only once something is filed, and only in that court. A sector-specific regulator’s guidance, where one exists, sits on top of the general privacy principles rather than replacing them — Alberta’s HIA guidance does not disapply PIPEDA’s general principles, it specifies what they mean for a health custodian.
A physiotherapy clinic’s in-house counsel is drafting a demand letter over a billing dispute and, separately, the clinic is piloting an AI scribe for patient notes. The demand letter engages the law-society layer (confidentiality of the file, competence in using the drafting tool, billing for time actually spent) and, if it becomes part of a court filing, whichever court’s AI notice applies. The AI scribe engages an entirely different layer: Alberta’s Health Information Act and the OIPC’s scribe-specific guidance, which has nothing to do with the law society or the courts. Treating AI use at the practice as one compliance question, answerable once, misses that these are genuinely separate obligations running in parallel. (treadstonelaw.ca, confidentiality clause remedies)
For all the structural difference between a court notice, a law society guideline and a privacy commissioner’s principles, three things recur in every one surveyed for this hub: a verification expectation (someone accountable has to check the output before relying on it), an accountability rule (the person or organization using the tool remains responsible, not the tool or its vendor), and a consistent refusal to ban the technology outright. The Law Society of Alberta’s survey makes the last point explicit across every body it reviewed — “none have attempted to ban its use.” (LSA, Gen AI Rules of Engagement) That pattern is itself informative: every regulator that has looked closely enough to write guidance has concluded the tool is usable, and the real work is in specifying how.
The cost of treating these layers as one question shows up at the worst possible time — after a filing has already gone in, or after a vendor contract has already been signed. Mapping the layers before adopting a tool means answering three separate questions once: which court or tribunal rules might apply if this work ends up filed somewhere; which professional body governs the individual doing the work; and which privacy statute governs the underlying data, including any sector-specific layer like Alberta’s HIA. Answering those three questions in advance is considerably cheaper than untangling them after a complaint, a filing challenge, or a regulator’s inquiry.
Related: AI in Canadian healthcare records, can lawyers use generative AI in Canada, and why professional bodies are cautious on AI
No. Courts, law societies, and privacy commissioners each regulate a different thing — what happens in a proceeding, how a licensed professional practises, and how personal information is handled — and none of the three has authority over the other two.
No, they operate on different levels. Professional-conduct guidance addresses how a licensee practises; privacy law addresses how personal information is collected, used and disclosed. A professional can meet one and still fall short of the other.
It varies by court, not by profession generally: Manitoba's and Yukon's 2023 practice directions require it; Alberta's tri-court notice does not; the Federal Court requires a Declaration only where AI content resembles that of a co-author in a specific filing.
A diligence review has to identify which of these levers is engaged before it can assess exposure.
