Employers using AI to support HR decisions should be prepared to defend those decisions under existing human rights law. AI can help employers make people management decisions faster, but it can also obscure how protected characteristics influence outcomes. Employers that rely on AI-assisted tools without understanding or documenting their use may expose themselves to discrimination claims, financial consequences, and reputational harm. Courts and governments are already responding, and their early interventions point to preventive steps employers can take now.
What is discrimination?
Human rights legislation across Canada protect employees from discrimination in hiring and throughout employment. Discrimination can arise whenever a decision about a term or condition of employment is based, even in part, on a protected characteristic, including Indigenous identity, race, colour, ancestry, place of origin, political belief, religion, marital status, family status, physical or mental disability, sex, sexual orientation, gender identity or expression, and age.
The proxy problem
Human rights law does not require an intention to discriminate.[1] If a protected characteristic is even a minor part of a decision not to hire or to terminate an employee, the employer may have violated the employee’s human rights. The risk is that ordinary employee information can act as a proxy for a protected characteristic. For example, an employee’s irregular hours may reflect their parental caregiving obligations. A termination decision that treats those hours as a negative factor may therefore be discriminatory.
A blind spot in AI-assisted decision-making
AI systems can intensify this proxy problem because they assign weight to data points and measure them against standards that users may not fully understand. A factor that is neutral for one employee may be discriminatory for another. This problem can be difficult to detect, particularly where systems learn from prior decisions and repeat the same flawed patterns.
Human review helps, but it is not a complete answer. Selective adherence—the tendency to accept AI outputs that align with existing biases and reject outputs that do not—means reviewers may fail to test AI-generated conclusions when scrutiny matters most. [2]
The legal response to AI-assisted decision-making
Canadian courts have not yet squarely addressed AI discrimination in employment, but immigration cases offer useful guidance. Applicants have argued that AI-assisted application review led to unfair or unreasonable decisions. The Courts’ response has been consistent: AI may affect decision-making, but that impact will not be legally recognized without specific evidence.
In Hagenshenas v Canada, the Federal Court held that AI use alone does not make a decision unreasonable. [3] In Raja v Canada, the Court held that expediency is not enough: even if AI allows an officer to spend only a “fraction of the time” on an application, specific evidence supporting a claim that the decision was unreasonable is still required. [4] The Court has also held that a decision-maker’s independent reasons is enough to establish what was considered, even when AI is relied on. [5]
These cases suggest three practical points for employers. First, AI use is not itself unlawful. Second, employees will need evidence linking AI-assisted decision-making to discrimination. Third, employers will be better positioned to respond if the human decision-maker can explain the independent reasoning behind the decision.
In addition to the Court’s evidence grounded approach, governments are also taking a disclosure-based approach. Ontario’s Working for Workers Four Act, in force January 1, 2026, requires Ontario employers to disclose AI use in screening, assessing, or selecting job applicants.[6] Nova Scotia’s Bill 234 proposes to go further by requiring disclosure of AI used for job postings, its specific purpose, and the applicant information provided to it.[7] These regimes may not create stand-alone penalties, but they may give employees evidence to support human rights claims.
The legal approach is still developing, but the warning is clear: AI-assisted HR decisions will not go unregulated, and employers should be ready to defend how these tools are used.
Where does this leave employers?
AI does not change the basic human rights analysis. Employees need only show that a protected characteristic was a factor in an employment decision.[8] That evidentiary burden may be easier to meet where disclosure laws reveal how AI was used and what information was provided to it. HR tools that assess risk, rank candidates, or evaluate performance may therefore create greater exposure than tools that simply organize information.
Employers and HR professionals should reduce that exposure by:
- Training staff on the specific risk presented by selective adherence;
- Mandating review and independent reasoning for all AI-assisted decisions to build the evidence needed to defend an AI discrimination claim; and
- Limiting AI use to the earlier stages of people management decisions to clearly demonstrate a break between AI outputs and the human judgement guiding final decisions.
The law is catching up with technology. Employers that treat AI governance as legal risk management today will be better positioned to defend their decisions tomorrow.
[1] See Ont. Human Rights Comm. v Simpsons-Sears, 1985 CanLII 18 (SCC), [1985] 2 SCR 536 at para 14 where the Supreme Court of Canada stated that intention should not be a governing factor in construing human rights legislation aimed at the elimination of discrimination; see also for example Human Rights Code, R.S.B.C. 1996, c. 210 at s. 2.
[2] Saar Alon-Barkat & Madalina Busuioc, “Human–AI Interactions in Public Sector Decision Making: ‘Automation Bias’ and ‘Selective Adherence’ to Algorithmic Advice” (2023) 33:1 J Pub Admin Research & Theory 153 at para 154.
[3] Hagenshenas v Canada, 2023 FC 464 at para 28 [Hagenshenas].
[4] Raja v Canada, 2023 FC 719 at para 28-30 [Raja].
[5] Mehrara v Canada (Citizenship and Immigration), 2024 FC 1554 at para 46-47 [Mehrara].
[6] Employment Standards Act, 2000, S.O. 2000, c. 41 at s. 8.4(1), as amended by the Working for Workers Four Act, 2024, c. 3, Sched. 2, s. 2(1).
[7] Bill No. 234, Helping Working People Get Ahead Act, 1st session, 65th General Assembly, Nova Scotia, 2026, s. 4(11C).
[8] Moore v British Columbia (Education), 2012 SCC 61 at para 33.
