Automated Decisions Under Law 25: Notice and Review Rights
Automated decisions under Law 25 carry specific duties. When a Quebec enterprise decides about a person based exclusively on automated processing, it must tell them, explain the information and principal factors used on request, and let them submit observations to a staff member who can review the decision. Here is how to comply.

Key takeaways
- Section 12.1 applies to decisions based exclusively on automated processing of personal information.
- Notice must be given no later than when the person is informed of the decision.
- On request, enterprises must explain the information used and the principal factors and parameters.
- Individuals can submit observations to staff who are in a position to review the decision.
- Penalties reach CAD 10 million or 2% of worldwide turnover under s. 90.12.
The rule on automated decisions under Law 25
Quebec was the first Canadian jurisdiction to legislate specifically on automated decision-making about individuals, and the rules on automated decisions under Law 25 are now a reference point for AI governance across Canada. Section 12.1 of the Act respecting the protection of personal information in the private sector, added by Law 25 and in force since September 2023, applies when an enterprise uses personal information to render a decision based exclusively on automated processing.
The enterprise must tell the person about it no later than when it informs them of the decision, must explain on request the information used and the reasons and principal factors behind the decision, and must give the person an opportunity to submit observations to a staff member who can review the decision. Rules on automated decisions under Law 25 therefore affect AI tools in hiring, credit, insurance, pricing, fraud detection and customer service whenever no human is meaningfully involved in the outcome.

This note explains when the rule applies, what notice and explanation must cover, how the review right works, how it interacts with other Law 25 obligations and with federal law, and how to design compliant processes. It is general information based on the Act updated to 15 October 2026, not legal advice.
When does section 12.1 apply?
Section 12.1 has three triggers. There must be a decision about a person; it must be rendered on the basis of personal information; and it must be based exclusively on automated processing of that information. The key interpretive question is the word "exclusively". A process where an algorithm produces a score that a human then reviews and can override, with genuine authority and information to do so, is generally outside the rule. A process where the algorithm's output is applied automatically, or where human sign-off is a formality, is likely within it.
Enterprises should not assume that any human involvement takes a process out of scope. A reviewer who sees only the system's recommendation, has no time or training to assess it and almost never departs from it may not change the character of the decision. Conversely, designing meaningful human review can be a legitimate compliance strategy, but it must be real and documented.
| Use case | Likely within s. 12.1? | Key compliance step |
|---|---|---|
| Applicant tracking system automatically rejects candidates below a score | Yes, if no human reviews rejections | Notice at rejection; explanation and review on request |
| Instant online credit decision | Yes, where approval or refusal is automatic | Notice with decision; factors explained on request |
| Fraud model flags a transaction for an analyst who decides | Usually no, if analyst review is genuine | Document human review criteria and authority |
| Dynamic pricing for an individual customer | Possibly, depending on whether it is a decision about the person | Assess whether pricing is individualized using personal information |
| Chatbot answering general questions | Generally no decision about the person | Consider transparency and profiling rules instead |
Notice and explanation duties
Under s. 12.1 of the Private Sector Act, the enterprise must inform the person concerned that the decision was based exclusively on automated processing no later than when it informs them of the decision. A rejection email, a credit refusal screen or a pricing notice should therefore include a short statement that the decision was made by automated means and how to request more information.
On request, the enterprise must also inform the person of the personal information used to render the decision, of the reasons and the principal factors and parameters that led to it, and of the right to have the personal information used corrected. The explanation must be meaningful to the individual. It does not require disclosure of source code or trade secrets, but it should identify the main inputs and how they influenced the outcome, for example that the application was refused primarily because reported income was below a threshold and the debt ratio exceeded a limit.
Explainability must therefore be built into system design. Models whose outputs cannot be explained at the level of principal factors make compliance difficult. Enterprises procuring AI tools should require vendors to provide the information needed to give these explanations, and should test that the explanations are accurate before the tool goes live.
The right to submit observations and obtain review
Section 12.1 also requires that the person be given the opportunity to submit observations to a member of the enterprise's personnel who is in a position to review the decision. This is a human review right: a staff member with authority to change the outcome must consider what the person says. The enterprise should define who reviews, what information they see, how quickly they respond and how the outcome is communicated.
Review processes should be proportionate but real. For high-volume decisions, a dedicated team with escalation criteria may be appropriate. The reviewer should be able to correct inaccurate data, consider context the system missed and override the decision where appropriate. Records of reviews and outcomes help demonstrate compliance and can reveal systematic problems in the model. If reviewers frequently overturn the system's decisions for the same reason, that is a signal to retrain or recalibrate the model, or to change the data on which it relies, rather than to keep correcting errors one complaint at a time.
Map every process in which a system, rather than a person, makes the final call about an individual. For each one, add an automated decision notice to the communication of the decision, prepare an explanation template listing the principal factors, and designate the staff who will handle observations and review requests.
Examples: AI in hiring and credit
Hiring is a common use case. Résumé screening tools, online assessments and video interview analytics can reject candidates automatically. Employers using such tools in Quebec should include an automated decision notice in rejection messages, be able to explain the main factors, and offer human review. Employers operating in Ontario must also, since 1 January 2026, disclose in publicly advertised job postings whether they use artificial intelligence to screen, assess or select applicants, under amendments to the Employment Standards Act, 2000 described in the province's guide to the Act. Human rights law adds another layer: automated tools must not discriminate on protected grounds, and employers remain responsible for vendors' tools. See our page on employment law.
In credit and insurance, automated underwriting is widespread. Financial institutions in Quebec should also consider the Autorité des marchés financiers' expectations on AI governance, discussed in our note on the AMF guideline on AI. Explanations of credit decisions should be consistent with consumer protection and credit reporting rules, and adverse decisions should be reviewable by staff with authority to change them.
Governance for automated decisions under Law 25
Compliance is easier when automated decision-making is governed centrally. Law 25 requires every enterprise to designate a person in charge of the protection of personal information, by default the person with the highest authority, and to adopt governance policies. Those policies should address automated decisions under Law 25 explicitly: who approves the deployment of a new decision system, what testing for accuracy and bias is required, how notices and explanations are maintained and how review requests are tracked. A simple register of decision systems, owners and review metrics allows the person in charge to oversee them and to answer questions from the CAI.
Testing should be ongoing rather than a one-time exercise. Models drift as data and conditions change, and a model that was accurate at launch can become less reliable or start to produce disparate outcomes for some groups. Periodic validation, monitoring of override rates in human review and analysis of complaints can reveal problems early. Where the enterprise relies on a vendor's model, the contract should give it access to the information needed for that validation.
Interaction with consumer protection and human rights law
Automated decisions under Law 25 can also raise issues under Quebec's Charter of human rights and freedoms, which prohibits discrimination in employment and in the provision of goods and services, and under consumer protection law where pricing or eligibility decisions are opaque or misleading. Explaining principal factors helps detect proxies for protected characteristics, such as postal codes or gaps in employment history that correlate with disability or family status. Enterprises should treat the explanation duty as an opportunity to audit fairness as well as a compliance obligation.
Related Law 25 obligations
Section 12.1 sits alongside other Law 25 rules that often apply to the same systems. Enterprises must carry out privacy impact assessments for projects to acquire, develop or overhaul information systems involving personal information, and those assessments are the natural place to evaluate automated decision risks. Law 25 also requires enterprises that use technology allowing a person to be identified, located or profiled to inform the person and, in some cases, to ensure such functions are deactivated by default. Consent, transparency and data minimization rules apply to the data used to train and run models, and transfers of data outside Quebec to AI vendors require a separate assessment; see our note on transfers outside Quebec.
Non-compliance can lead to CAI orders and administrative monetary penalties of up to CAD 10 million or 2 per cent of worldwide turnover under s. 90.12, and to penal fines under s. 91. Individuals may also claim damages.
Federal and other provincial context
There is no federal statute equivalent to s. 12.1 in force. The federal Bill C-27, which would have enacted the Consumer Privacy Protection Act with automated decision provisions and the Artificial Intelligence and Data Act, died when Parliament was prorogued in January 2025, and its successor privacy bill, C-36, had not received royal assent as of November 2026. Organizations subject to PIPEDA must still meet its openness, accuracy and accountability principles when using automated tools. Companies operating nationally should therefore use Quebec's rules as a practical baseline. Our artificial intelligence page tracks federal and provincial developments.
Compliance checklist for automated decisions under Law 25
- Inventory systems that make decisions about individuals without meaningful human involvement.
- Decide, for each, whether to add genuine human review or to comply with s. 12.1.
- Add automated decision notices to the communication of decisions.
- Prepare explanation templates covering data used and principal factors.
- Designate and train staff to receive observations and review decisions.
- Require vendor cooperation on explainability, accuracy and testing.
- Integrate automated decision risks into privacy impact assessments.
How TRW & Co helps
We advise employers, lenders, insurers, retailers and technology companies on automated decisions under Law 25 and related AI governance. We map automated decision processes, draft notices and explanation templates, design review procedures, negotiate AI vendor contracts and integrate automated decision risks into privacy impact assessments. Where a matter requires a lawyer licensed in Quebec, we work with Quebec counsel. To discuss your systems, contact us.
Frequently asked questions
What counts as a decision based exclusively on automated processing?
It is a decision about a person, made using personal information, where no human meaningfully participates in the outcome. If a staff member genuinely reviews the system’s output and has the authority and information to change it, the decision is generally not exclusively automated. Formal sign-off without real review may not be enough.
When must the person be told about an automated decision?
Under s. 12.1, the enterprise must inform the person that the decision was based exclusively on automated processing no later than when it informs them of the decision itself. In practice, the notice should be included in the message communicating the decision.
What explanation must be provided?
On request, the enterprise must tell the person 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. The explanation must be meaningful but does not require disclosing source code.
Do automated hiring tools fall under s. 12.1?
They can. If an applicant tracking system or assessment tool rejects candidates without meaningful human review, the employer must give notice, explain the principal factors on request and allow review by staff. Human rights laws also apply, and Ontario separately requires disclosure of AI use in publicly advertised job postings.
Is there a federal equivalent to Quebec’s automated decision rule?
Not currently. Federal Bill C-27, which contained automated decision provisions, died on prorogation in January 2025, and the successor privacy bill, C-36, had not received royal assent as of November 2026. PIPEDA’s general principles still apply to organizations under federal jurisdiction.
Sources
General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.


