No one sets out to build a discriminatory hiring tool. The problem is that a hiring tool does not need to be told to discriminate — it only needs to be trained on data that already reflects a pattern, and it will faithfully reproduce that pattern at scale, in a decision that looks perfectly neutral on its face.
Key takeaways
Canada’s federal, provincial and territorial privacy commissioners’ joint AI principles state the mechanism directly: developers must evaluate training data “to ensure that they do not replicate, entrench, or amplify historical or present biases – or introduce new biases,” because failing to do so “may be more likely to result in discriminatory outcomes based on race, gender, sexual orientation, disability, or other protected characteristics, particularly where they are used as part of an administrative decision-making process… or in highly impactful contexts such as… employment.” OPC, generative AI principles A resume-screening or candidate-ranking tool trained on a company’s own historical hiring data does not need anyone to instruct it to favour one group — it only needs that history to already skew that way, which most hiring histories, to some degree, do.
The mechanism has a specific name in Canadian law, and it does not require intent. The Canadian Human Rights Act defines a discriminatory practice as one that “deprives or tends to deprive an individual or class of individuals of any employment opportunities on a prohibited ground of discrimination” — language broad enough to catch a hiring tool that produces a skewed outcome even though no one built it to discriminate. Canadian Human Rights Act, s.10 That federal Act, like PIPEDA, binds only federally regulated employers directly; a provincially regulated employer instead answers to its own province’s human rights code, most of which use comparable language and are enforced through a dedicated tribunal rather than a privacy regulator. Treadstone Law, HRTO vs. court for workplace discrimination in Ontario
Québec’s privacy regulator, the CAI, publishes the most concrete Canadian guidance specifically on AI hiring bias. Its recruitment guidance states that the employer “doit porter une attention particulière au critère de nécessité, à la transparence et aux biais discriminatoires des algorithmes” — must pay particular attention to necessity, transparency, and the discriminatory biases of the algorithms — and requires a privacy impact assessment (évaluation des facteurs relatifs à la vie privée) before an AI hiring system is used at all. CAI, AI in recruitment (Québec) Where a hiring decision is based exclusively on automated processing, the employer must notify affected candidates no later than when it informs them of the decision, provide further information on request, and offer the right to have the decision reviewed. The same guidance singles out emotion- or psychological-state-recognition systems used in video interviews as “très peu susceptibles d’être proportionnels aux besoins de l’employeur” — very unlikely to be proportionate to the employer’s needs — and treats that use, in the great majority of cases, as an inappropriate use of AI. The CAI also requires that the employer’s organization have reached the necessary technological maturity to use the tool responsibly, with staff who are adequately trained and understand the limits of the tools they rely on — a bias problem is, in part, a competence problem, in the CAI’s own framing.
Ontario employers with 25 or more employees must disclose, in any publicly advertised job posting, “a statement disclosing the employer’s use, if any, of artificial intelligence to screen, assess or select applicants for the position.” Ontario ESA, job posting requirements That is a transparency requirement — it tells a candidate a tool was used, not whether it was fair. It solves a different problem than Québec’s bias-and-review regime, and the two should not be treated as equivalents. For the disclosure rule’s full mechanics — the threshold, the timing, what counts as a posting — see Ontario’s AI disclosure rule for job ads.
There is no federal Canadian statute addressing AI discrimination in hiring specifically. What exists federally is the OPC’s joint AI guidance quoted above — non-binding privacy principles, not enforceable anti-discrimination law — alongside the OPC’s general recommendation that impacted individuals be given “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 Canadian employer using an AI hiring tool outside Québec is relying substantially on that non-binding guidance and on Ontario-style disclosure rules, not on a dedicated federal bias regime.
A company’s resume-screening tool was trained on ten years of its own past hires, which happened to skew heavily toward one demographic for reasons that had nothing to do with candidate quality — an old sourcing channel, a past hiring manager’s preferences. The tool will reproduce that skew as a scoring pattern, and nothing about its output will announce that this is what happened; it will simply rank candidates who resemble the historical pattern more highly. Before deploying such a tool, the practical checks are: has anyone reviewed whether the training data over-represents a particular group for reasons unrelated to merit; is there a human review point before a candidate is rejected, consistent with the OPC’s challenge-mechanism principle; and, if the employer operates in Québec, has the required privacy impact assessment actually been completed before go-live, not after a complaint.
Related: Ontario’s AI disclosure rule for job ads, is AI resume screening legal in Canada, and AI and employee privacy at work.
Yes, mechanically. A tool trained on historical hiring data that already skews toward one group will reproduce that skew as a scoring pattern, without anyone having instructed it to. Canada’s privacy commissioners name this risk directly for employment specifically.
Yes. Québec requires a privacy impact assessment before an AI hiring system is used, and a notice-and-review right for exclusively automated hiring decisions. Ontario’s rule is a disclosure duty in job postings for employers with 25 or more employees — a different, narrower requirement.
No. There is no federal statute addressing this specifically. What exists federally is non-binding privacy guidance from the OPC, alongside Québec’s regulator-enforced regime and Ontario’s disclosure rule, which sit at the provincial level.
Usually, but not only. Québec’s regulator also treats staff competence as part of the problem — an organization using a tool without the technical maturity to understand its limits is itself named as a risk factor, separately from whatever is wrong with the data.
A short call is enough to walk through what a defensible pre-launch review should actually check.
