A manager asks an AI tool to draft a performance review from months of emails, chat logs, and project notes. Nothing about that changes the legal rules that already governed performance reviews — it just means those rules now apply to a document a person did not fully write.
Key takeaways
Personal information an employer holds and uses about a staff member — including whatever an AI tool draws on to write a review — is subject to PIPEDA’s Schedule 1 accuracy principle: “personal information shall be as accurate, complete, and up-to-date as is necessary for the purposes for which it is to be used.” PIPEDA, Schedule 1, cl.4.6, Accuracy An AI tool summarizing months of communications into a review is exactly the kind of use that principle was written for — the summary has to reflect what actually happened, not merely what the underlying text pattern suggested, and an employer relying on an AI-drafted summary has not been relieved of that duty by delegating the drafting.
Canada’s privacy commissioners’ joint AI principles are unambiguous on this point: organizations using generative AI should “know that accountability for decisions rests with the organization, and not with any kind of automated system used to support the decision-making process,” and should “ensure that impacted individuals are provided with an effective challenge mechanism for any administrative or otherwise significant decision made about them… and allowing them the opportunity to request human review.” OPC, generative AI principles A performance review that materially affects someone’s standing, pay, or continued employment is squarely the kind of “significant decision” this principle addresses. An AI tool can draft the language; it cannot be the thing that decided the outcome, and the organization cannot point to the tool as the source of an unfair result.
In Québec, that human-review expectation has a specific statutory source and goes further than the OPC’s general principle. The province’s private-sector privacy law requires that anyone who bases a decision exclusively on automated processing of personal information “inform the person concerned accordingly not later than at the time it informs the person of the decision,” disclose on request the information used and the reasons and principal factors that led to it, and give the person “the opportunity to submit observations to a member of the personnel of the enterprise who is in a position to review the decision.” Act respecting the protection of personal information in the private sector, s.12.1 That right applies specifically where the review was exclusively automated — a manager who reads an AI-drafted review before relying on it sits outside it entirely.
If an AI-informed review is used as the basis for a significant, unilateral change to someone’s employment — a demotion, a substantial pay cut, a stripping of key responsibilities — Ontario’s existing constructive dismissal doctrine applies regardless of what produced the review. As Treadstone Law’s guidance on this doctrine explains, “an employer cannot make fundamental changes to an employee’s job terms without consent. If they do — and the change is significant enough — it can constitute constructive dismissal.” Treadstone Law, changing job conditions without consent The test looks at whether the change itself was substantial and unilateral, not at what tool the employer used to justify it. An AI-drafted review that led to the change is evidence in that analysis; it is not a separate legal category.
PIPEDA’s Schedule 1 individual access principle gives an employee a direct route into whatever an AI tool drew on: “upon request, an individual shall be informed of the existence, use, and disclosure of his or her personal information and shall be given access to that information. An individual shall be able to challenge the accuracy and completeness of the information and have it amended as appropriate.” PIPEDA, Schedule 1, cl.4.9, Individual Access Applied to an AI-drafted review, that means an employee who disputes what the tool summarized is not limited to disputing the conclusion informally — they can request the information the review was built from and challenge its accuracy and completeness directly, which is a materially stronger position than simply asking a manager to reconsider.
The three legal layers above set a floor, not a template for how to actually run an AI-assisted review process well. In practice, that means keeping a record of what data the AI tool drew on for a given review, so an accuracy challenge under PIPEDA’s access principle can actually be answered; having a named person — not a system — sign off on the review before it is delivered, consistent with the OPC’s accountability principle; and giving the employee a genuine opportunity to respond to the review’s content before any decision tied to it becomes final, which is both good practice and the kind of “challenge mechanism” the OPC’s guidance specifically asks for.
None of these three layers — PIPEDA’s accuracy principle, the OPC’s human-accountability principle, or Ontario’s constructive dismissal test — was written with AI in mind, and none of them is a standalone “AI performance review law.” They are general rules that already applied to the employment relationship, now being applied to a new tool that changes how the underlying document gets produced, not what legal test governs its consequences.
An AI tool drafts a review summarizing months of an employee’s chat and email activity and flags them as underperforming, and the employer demotes them on that basis. Three questions actually matter: was the underlying data the summary drew on itself accurate and up to date, as PIPEDA requires; did a person actually review and decide on the outcome, rather than treating the AI’s flag as the decision itself; and, separately from either of those, was the resulting demotion itself a substantial enough change to meet Ontario’s constructive dismissal test. An employer can pass the first two and still lose on the third — and can fail the first two even where the resulting change was modest enough to survive the third. They are independent questions, not one combined AI test.
Related: can employers monitor staff with AI, AI and employee privacy at work, and can staff be disciplined for using AI.
An AI tool can draft or inform the review, but Canada’s privacy commissioners are explicit that accountability for the resulting decision stays with the organization, not the tool. A person still has to own the outcome.
There is no general Canadian right to this outside Québec, where an exclusively automated decision triggers a notice-and-review requirement. Federally, the OPC’s principles recommend that significant automated outputs be meaningfully identified, but this is guidance, not an enforceable right, outside Québec.
No. The same constructive dismissal test applies regardless of what produced the paperwork behind the change — what matters is whether the change itself was substantial and unilateral.
A short call is enough to check the accountability and documentation steps that actually protect the business.
