Can AI help you hire without discriminating?

Hiring is one of the places where AI is most tempting and most risky. A single job posting can draw hundreds of applications, and a tool that reads them all in seconds looks like a gift to a small team. The risk is that a model trained on past hiring decisions, or on text from the web, can learn the patterns hidden in that history, for instance favouring certain schools, neighbourhoods or career paths that track gender, origin or age. The output looks neutral because it comes from a machine, which makes the problem harder to see.

The answer is not to avoid AI entirely but to give it a narrow job. AI is good at summarizing, extracting and organizing. It should not be the one deciding who gets an interview. In 2026 that is also the direction the law is taking on both sides of the border.

Where bias gets in

Three entry points cover most of the trouble:

Postal codes deserve a special word. They look harmless but often stand in for origin or income. Illinois now names this directly: employers there may not use zip codes to stand in for protected characteristics.

A worked screening workflow

Here is a process a 30-person company in Ottawa or Gatineau could run with an ordinary AI assistant on a business plan:

1. Write the criteria. For an office coordinator: French and English at a working level, two years of scheduling or administrative work, comfort with spreadsheets, availability on site three days a week.
2. Strip identifiers. Remove names, photos, addresses, dates of birth and graduation years from the copies the AI reads.
3. Ask for evidence, not scores. Use a prompt like the one below.
4. A person decides. The hiring manager reads every summary, opens the full file for anyone marked "unclear," and records the reason for each decision.
5. Check outcomes. After the round, compare how many people from different groups who self-identified passed each stage. A large gap is a signal to review the criteria or the tool.

You are helping a hiring manager organize applications. You do not decide.
Job criteria:
1. Working French and English
2. Two years of scheduling or administrative experience
3. Comfortable with spreadsheets
4. Available on site three days a week
For the application below, list each criterion and quote the passage
that shows it is met, not met, or unclear. Do not guess. Do not comment
on age, gender, origin, health, family situation or any personal trait.
Do not rank or score the candidate.
[application text]

Which rules apply to AI in hiring in Quebec, Ontario and the US?

Quebec. The private-sector privacy act, as amended by Law 25, does not ban AI in hiring, but section 12.1 applies when a decision about a person rests exclusively on automated processing of their personal information. The employer must tell the candidate, at the latest when it informs them of the decision. On request, it must explain the personal information used, the main reasons and factors behind the decision, and the right to have the information corrected, and the candidate must be able to present observations to a staff member who can review the decision. Keeping a person genuinely in charge of each decision, as in the workflow above, keeps you out of the "exclusively automated" category, but section 8 still requires telling applicants why and how you collect their information.

Ontario. Since January 1, 2026, the Employment Standards Act requires employers with 25 or more employees to say in each publicly advertised job posting whether they rely on AI to screen, evaluate or pick applicants. Every employee counts, part-time included, across all Ontario locations. A short sentence is enough, such as "This employer uses artificial intelligence to help screen applications," and the duty stays with the employer even when a recruiting firm does the screening. The definition of AI is broad, so an applicant-tracking system that ranks résumés may well be covered. The same postings must also include expected pay, say whether the vacancy exists, and avoid Canadian-experience requirements.

Illinois. From January 1, 2026, the Illinois Human Rights Act makes it a civil rights violation for an employer to use AI in recruitment, hiring, promotion, discipline or other employment decisions in a way that has the effect of discriminating on protected grounds. Intent is not required; the effect is enough. Employers must also give notice when they use AI for these purposes, with details left to rules from the state's Department of Human Rights.

New York City. Under Local Law 144, enforced since July 2023, an automated employment decision tool may be used only if it had a bias audit within the past year, with a public summary of the results and a notice to candidates at least ten business days before use.

If you hire in Europe, AI used to recruit or select people is on the EU AI Act's high-risk list; after the 2026 Omnibus, those rules apply from December 2, 2027. For a Canadian business hiring across several of these places, the safest common denominator is the same: disclose AI use, keep a person in charge, record your reasons, and test for unequal results. Before rolling a tool out widely, have an employment lawyer review your postings and notices for each place you hire.