AI screening and human rights law in Canada: where the risk lies
Human rights law does not mention AI, but it already covers what a screening tool does. Where the risk sits: proxy criteria, past hiring data, formats some people cannot use, and unchecked results.
6 minute readAI in hiring
Ontario’s Human Rights Code does not mention artificial intelligence, and it does not need to. It gives every person a right to equal treatment in employment without discrimination on listed grounds, it forbids infringing that right directly or indirectly, and it treats a neutral-looking requirement that excludes a protected group as discrimination unless the requirement is reasonable and bona fide. A screening tool applies requirements to every application, quickly and at scale. The legal risk sits in what the tool is told to look for, what it can infer, who it cannot serve, and whether anyone checks what it produces.
This article uses Ontario’s Code as the reference. If you hire in another province, or your organization is federally regulated, check the human rights statute that applies there.
The parts of the Code that matter for screening
- The grounds. Section 5(1) gives a right to equal treatment in employment without discrimination because of race, ancestry, place of origin, colour, ethnic origin, citizenship, creed, sex, sexual orientation, gender identity, gender expression, age, record of offences, marital status, family status or disability. For this purpose, age means 18 or older.
- Directly or indirectly. Section 9 says no person shall infringe a Code right, or do, directly or indirectly, anything that infringes it.
- Neutral rules with unequal effects. Under section 11, a requirement, qualification or factor that excludes, restricts or prefers a group identified by a ground infringes the right unless it is reasonable and bona fide. It cannot be found reasonable and bona fide unless the group’s needs cannot be accommodated without undue hardship.
- Postings, forms and questions. Sections 23(1) and 23(2) cover advertisements, application forms and written or oral inquiries that directly or indirectly classify or indicate qualifications by a ground.
- Agencies. Section 23(4) says an employment agency infringes the right when it discriminates in receiving, classifying, disposing of or otherwise acting on applications, or in referring applicants to an employer.
Read together, these describe most of what a screening tool does. Its criteria are requirements and qualifications. Its scoring classifies applicants. Its rejections dispose of applications, and for an agency, its shortlist is a referral. None of these sections makes an exception for work done by software.
Where the risk lies
Criteria that stand in for a ground
A tool can be told to prefer recent graduates, applicants with no gaps in their work history, or experience gained in a particular country. None of these names a ground, and each can track one: age, family status or disability, place of origin. In its February 2024 submission on Bill 149, the Ontario Human Rights Commission noted that AI technologies have used personal information such as names, postal codes and gaps in employment history to make inferences about race, disability, age and other Code grounds.
The test for every criterion is the one in section 11. Is it reasonable and bona fide for this job, and could the people who do not meet it be accommodated?
Training on yesterday’s hires
A tool that learns what a good hire looks like from past hires learns past preferences along with them. The OHRC’s submission cites a review of the Peel District School Board, which found that an algorithmic tool used to vet teaching candidates had screened out qualified racialized candidates. The tool had been designed to mirror previous successful hires. Ask any vendor what the tool was trained on, and whether it adjusts itself to your own past decisions.
Formats some candidates cannot use
The same submission warns that interview technologies may be less reliable for applicants with a speech impediment, applicants who use a screen reader and applicants whose first language is different, and that chatbots may not respond properly to a request for accommodation. Under section 11(2), the possibility of accommodation is part of deciding whether a requirement is reasonable. A process that offers one way to apply, and no way to reach a person, makes that duty hard to meet.
Results nobody checks
A tool can apply the same rules to everyone and still produce unequal results. The IPC and OHRC principles released in January 2026, which are addressed to Ontario’s public sector and broader public sector, say so directly: using an AI system uniformly with diverse groups may seem neutral and still result in adverse impact discrimination. They expect those institutions to identify and address systemic discrimination in the design and deployment of AI systems, and the same reasoning applies to any employer under section 11. The Human Rights AI Impact Assessment, released by the OHRC and the Law Commission of Ontario in November 2024, gives a two-part method for this: Part A assesses a system for human rights implications, and Part B works through mitigation.
A worked example
Take a hypothetical agency that receives a job order for a warehouse shift lead. The client’s description asks for “3 to 6 years of experience”, and the agency’s screening tool turns that into two rules: applications with less than three years of shift leadership are scored down, and so are applications with more than six years of experience, labelled “overqualified”.
- The minimum has a link to the work. Three years of leading a shift is a requirement the agency can explain.
- The cap does not. It excludes people because they have worked longer, which in practice tends to mean older applicants. Under section 11, a rule like that has to be shown to be reasonable and bona fide.
- Section 23(4) applies to the agency’s own handling of applications, whatever the client wrote.
The fix is to keep the minimum, drop the cap and write the requirement as a skill: “at least three years leading a shift in a warehouse or distribution centre”.
Questions to ask before you switch a tool on
- Which criteria does the tool apply, and can we see and change them for each job?
- What information is removed before scoring, and can a recruiter see what was removed?
- Has the tool been tested for different results across groups, and can we see the method and the results?
- How does a candidate ask for an accommodation or for a person, and who receives the request?
- What is kept for each recommendation, so we can explain a decision later?
Our approach
TalentInitiator checks every scoring rubric before it can be approved and rejects criteria that act as proxies for a protected ground, such as age, graduation recency, employment gaps or Canadian experience. Its adverse-impact report compares selection rates across groups from people’s decisions, using voluntary self-identification kept apart from scores. The details are on our AI use page.
How to read this post
This post explains the rules in general terms, as they stood on the date above. It is not legal advice for your organization.
Sources
- Human Rights Code (Ontario)
- OHRC submission on Bill 149 (AI in hiring) Published February 13, 2024
- IPC and OHRC principles for the responsible use of AI Published January 21, 2026
- Human Rights AI Impact Assessment (OHRC and LCO) Published November 6, 2024