If you recruit in Quebec and use AI anywhere in the funnel, one provision deserves a bookmark: section 12.1 of the Act respecting the protection of personal information in the private sector, added by Law 25 (formerly Bill 64) and in force since 22 September 2023. It governs decisions based exclusively on automated processing of personal information. For Quebec Law 25 AI hiring, that single word, "exclusively," does a lot of work.
This guide explains what section 12.1 asks of employers, why a meaningful human decision changes the picture, and how to answer a candidate who asks "why?". You'll also find a sample notice in English and Canadian French.
#What does section 12.1 of Quebec Law 25 require?
Torys LLP's summary of the provision breaks it into a handful of duties. When an organization uses personal information to make a decision based exclusively on automated processing, it must:
- Inform the person that the decision was made that way, no later than the moment it tells them the decision.
- On request, tell them what personal information was used, the reasons and principal factors and parameters that led to the decision, and that they have the right to have the information corrected.
- Give them a chance to submit observations to a member of staff who is in a position to review the decision.
Torys notes these rights apply both to people outside the organization (customers, applicants) and inside it (employees). A general notice buried in terms and conditions, covering unspecified purposes, isn't enough. Describe the actual process.
#What counts as an "exclusively automated" hiring decision?
Picture the extremes first.
Clearly exclusive: Imagine a Montréal fintech receives 2,000 applications for a junior analyst role. A screening tool scores each résumé and automatically sends a rejection to everyone below a cut-off. No person looks at those files. That's a decision based exclusively on automated processing, and section 12.1 applies to each rejected applicant.
Clearly not exclusive: The same fintech runs structured interviews. An AI tool scores answers against a rubric and highlights the quotes that support each score. A recruiter reads the evidence, adjusts scores where the AI missed context, and decides who advances. A person made that decision.
The hard cases sit in between. Torys frames the inventory question as whether a process has "no human input on the outcome." That raises an obvious follow-up: what if there's a human, but only on paper?
#Why does a meaningful human decision matter?
Because a rubber stamp probably isn't a decision. If a recruiter clicks "approve" on 300 AI recommendations in four minutes, without opening a single answer, it's hard to argue the outcome had real human input. A cautious reading of section 12.1 treats that as automated in substance.
A meaningful human decision tends to look like this:
| Signal | Meaningful | Rubber stamp |
|---|---|---|
| What the reviewer sees | The candidate's answers and the evidence behind each score | A single score or a green/red label |
| Ability to disagree | Can override, and the system records it | Override is hidden, discouraged or never used |
| Time and attention | Enough to read the evidence | Bulk approval |
| Accountability | A named person owns the decision | "The system decided" |
| Training | Reviewers know the rubric and the tool's limits | None |
This is where design matters. MSZ Hire is built so that a human makes every decision: the AI scores each answer against the employer's rubric and quotes the part of the answer that supports the score, and a recruiter reviews that evidence before anyone advances or is declined. That's designed to support a genuine human decision under section 12.1. It doesn't remove your other Law 25 duties, such as consent and transparency, and it doesn't change Ontario's separate rule.
#Sample candidate notice for AI interviews in Quebec
Even if your process keeps a human in charge, telling candidates how AI is used builds trust and covers you if a step turns out to be more automated than you thought. Here's a notice you can adapt.
#English
About this interview. This interview is conducted by an AI interviewer that asks every candidate the same structured, job-related questions, answered by voice on camera. The interview is recorded so our recruitment team can review it, and the recording is deleted after [180 days]. An AI tool scores your answers against criteria set by [Employer] and highlights the parts of your answers that support each score. It does not analyze your face, voice tone or emotions. A member of our recruitment team reviews your answers and the scores, and a person makes every decision about your application.
You may ask us what personal information was used in assessing your application, the main reasons and factors behind a decision, and to correct any inaccurate information. You may also submit comments to a person who can review the decision. Contact: [privacy@employer.ca]. For accommodations or another interview format, contact [recruiting@employer.ca].
#Version française (Canada)
À propos de cette entrevue. Cette entrevue est menée par un intervieweur IA qui pose à chaque candidat les mêmes questions structurées liées au poste, auxquelles vous répondez à l'oral, devant la caméra. L'entrevue est enregistrée pour que notre équipe de recrutement puisse l'examiner, et l'enregistrement est supprimé après [180 jours]. Un outil d'IA évalue vos réponses selon des critères établis par [Employeur] et met en évidence les passages qui appuient chaque note. Il n'analyse ni votre visage, ni le ton de votre voix, ni vos émotions. Un membre de notre équipe de recrutement examine vos réponses et les notes, et toute décision concernant votre candidature est prise par une personne.
Vous pouvez nous demander quels renseignements personnels ont été utilisés pour évaluer votre candidature, les principales raisons et les principaux facteurs ayant mené à une décision, et faire corriger tout renseignement inexact. Vous pouvez aussi présenter vos observations à une personne en mesure de réviser la décision. Contact : [vieprivee@employeur.ca]. Pour une mesure d'adaptation ou un autre format d'entrevue : [recrutement@employeur.ca].
Note the Quebec usage: entrevue, not entretien. Small detail, big signal to candidates that you wrote this for them.
#How do you answer a candidate's request for reasons?
Sooner or later a candidate will write: "Why wasn't I selected, and was it a computer?" Here's a step-by-step that works whether or not section 12.1 technically applies.
- Acknowledge quickly. Confirm you received the request and tell the candidate who is handling it. Your privacy officer should be in the loop.
- Confirm identity. Make sure you're sending personal information to the right person, using the contact details on file.
- Pull the record. Gather the personal information used: application, interview transcript, rubric, scores, evidence, reviewer notes and the final decision.
- Explain the reasons and principal factors. In plain language, name the criteria that mattered most and how the candidate's answers related to them. For example: "The role required experience reconciling accounts. Your answers to questions 2 and 4 described general bookkeeping but not reconciliation." Avoid jargon such as "model confidence."
- Describe the human role honestly. Say who reviewed the application and what the AI tool did. If any step was fully automated, say so.
- Offer correction. If the candidate says something is wrong (a misheard answer, an outdated résumé), correct it and note the change.
- Invite observations and review. Give the name or role of a person who can review the decision, and actually have them look again.
- Document everything. Keep the request, your response and any change of outcome. Patterns in requests are useful feedback on your rubric.
#What else does Law 25 expect from AI hiring tools?
Section 12.1 is one piece. Torys also recommends that organizations inventory their automated processes, rank them by risk (data sensitivity, volume, impact on the person), document how each one works, and test for bias. If you're building that inventory across provinces, our plain-language map of PIPEDA, Bill C-36 and provincial privacy laws shows where Quebec fits.
Practical features that help with Quebec hiring in MSZ Hire include Canadian data hosting in AWS Montréal (ca-central-1), interviews in English and Canadian French, and Quebec-specific notices and consent shown before the interview starts.
#Your Quebec Law 25 AI hiring checklist
- Map every step where software makes or shapes a decision about a candidate.
- Mark any step with no human input on the outcome. Section 12.1 applies there.
- For human-reviewed steps, make the review real: evidence, override, accountability.
- Publish a clear notice in French and English.
- Prepare a template response for requests for reasons.
- Train recruiters on the process, and test it with a mock request.
Want to see what a structured, bilingual AI interview looks like before you write your notice? Try the candidate experience yourself. And when you're ready to run AI interviews in Quebec with a person making every decision, quoted evidence behind every score and data hosted in Montréal, MSZ Hire is ready when you are.
This article is general information, not legal advice. Consult a Quebec privacy lawyer about your specific situation.