DataPacts

Quebec Law 25 and AI Training Data

Checked

At a glance

Quebec Law 25 is the 2021 statute (S.Q. 2021, c. 25, formerly Bill 64) that rewrote Quebec's Act respecting the protection of personal information in the private sector, and for AI training data it decides four things: whether a new training use needs fresh consent, when a dataset is anonymized and leaves the Act, what may be shown to a prospective buyer, and what must happen before personal information leaves Quebec.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

The Act is CQLR c. P-39.1 as amended by Law 25, read in the Commission's administrative version as of 2026-11-17, together with the 2024 anonymization regulation. This page sets out the Quebec rules in AI training data licensing in the order a seller meets them.

Each rule here is dated and checked against the Quebec texts. The method is set out on the methodology page.

Law 25 in Force: Commencement Dates and the Duties a Training-Data Deal Meets

Law 25 came into force in stages, and every core duty a training-data deal meets today has applied since September 22, 2023, while portability, the anonymization regulation and its register came later. The table lists the six dates as of 2026-11-17.

Date What took effect Section (P-39.1 unless stated) Role in a training-data deal Source
2021-09-22 Law 25 assented (Bill 64); a few provisions in force Law 25 s.175(1) starting point of the staged timeline cai.gouv.qc.ca (CAI_Loi_25.pdf)
2022-09-22 person in charge of privacy protection; commercial-transaction exception; research communication with an assessment ss.3.1, 18.4, 21 sharing samples with a prospective buyer CAI administrative consolidation
2023-09-22 most duties: consent per purpose, use limits, de-identified and anonymized information, section 17 transfers, system assessments, penalties ss.3.3, 12 to 14, 17, 23, penalties the core of every deal CAI_Loi_25.pdf s.175; consolidation
2024-05-30 anonymization regulation in force (O.C. 783-2024 of 2024-05-01, published 2024-05-15) regulation ss.1 to 8, 10 when a set stops being personal information publicationsduquebec.gouv.qc.ca
2024-09-22 portability: information collected from a person returned in a structured, commonly used format s.27, third paragraph requests from people in a licensed set CAI_Loi_25.pdf s.175(3)
2025-01-01 anonymization register regulation s.9 record kept for each anonymized set publicationsduquebec.gouv.qc.ca

The regulation added the section 23 criteria in May 2024, and its register began on January 1, 2025.

Under section 14 of the private-sector Act, consent must be manifest, free and enlightened, given for specific purposes and requested for each purpose in simple, clear terms, so consent collected for a service does not by itself cover licensing the same records for AI training.

Consent lasts only as long as the purposes need, a written request must be presented separately from other information, and consent that does not meet these conditions has no effect.

Section 12 limits use to the purposes of collection unless the person consents, with express consent for sensitive information (medical, biometric or otherwise intimate, or sensitive by context).

Section 13 bars communicating personal information to a third party without consent unless the Act provides for it, again with express consent for sensitive information. The contract-side checks before licensing sit on the page about customer consent before licensing data for AI.

Section 12 lists five cases in which a business may use personal information for another purpose without consent, and none of them names AI training.

  • A compatible purpose, meaning one with a relevant and direct link to the original purposes; commercial or philanthropic prospecting is never compatible.
  • A use clearly for the benefit of the person concerned.
  • Fraud prevention or security.
  • Supplying a product or service the person requested.
  • Study, research or statistics, if the information is de-identified.

Whether training a model is a compatible purpose or "research" under the fifth case is not settled by any source read for this page as of 2026-11-17. Section 21 separately lets a business communicate information for study, research or statistics after a privacy impact assessment, in force since 2022-09-22.

De-identified Is Still Personal Information, Anonymized Is Not: Section 23

Quebec law separates two states of a dataset: information is de-identified when it no longer identifies the person directly and remains personal information under the Act (s.12), while it is anonymized only when it is at all times reasonable to expect, in the circumstances, that it irreversibly no longer allows the person to be identified directly or indirectly (s.23).

A business that uses de-identified information must take reasonable measures to limit the risk of re-identification (s.12).

When the purposes of collection are fulfilled, section 23 requires the business to destroy the information or to anonymize it for use for serious and legitimate purposes, following generally accepted best practices and the criteria set by regulation.

The Commission d'accès à l'information states that anonymized information is no longer personal information and may be used, communicated and kept without further obligation. The page on what anonymization and pseudonymization achieve legally compares the two states across jurisdictions.

The 2024 Anonymization Regulation: Steps, Criteria and the Register

Quebec's anonymization regulation (Règlement sur l'anonymisation des renseignements personnels, Order in Council 783-2024 of May 1, 2024) sets the criteria section 23 refers to and binds every business subject to the private-sector Act.

It was published in the Gazette officielle du Québec, Part 2, on 2024-05-15 and entered into force on the fifteenth day after publication, 2024-05-30, except the register (s.9), in force since 2025-01-01. The table lists the seven steps in the regulation's order, as of 2026-11-17; the analysis need not show zero risk, but the residual re-identification risk must be very low.

Step What the regulation requires Section Source
1 set the purposes for the anonymized information before starting; check them against s.23 again before any new purpose s.3 Gazette officielle
2 carry out the process under the supervision of a competent person s.4 Gazette officielle
3 remove direct identifiers, then run a preliminary re-identification risk analysis using the individualization, correlation and inference criteria and other reasonably available information, including public data s.5 (definitions s.2) Gazette officielle
4 choose techniques that follow generally accepted best practices, plus reasonable protection and security measures s.6 Gazette officielle
5 final risk analysis showing irreversible non-identification; zero risk not required, residual risk very low, judged on the circumstances and purposes, the nature of the information, the three criteria, available information, and the effort, resources and expertise re-identification would take s.7 Gazette officielle
6 periodic re-assessment, updating the last analysis for technological advances; if it fails, the data is no longer anonymized s.8 Gazette officielle
7 register: description, purposes, techniques and measures, dates of the analysis and any update (from 2025-01-01) s.9 Gazette officielle

The three criteria in section 2 are individualization (a person cannot be isolated), correlation (datasets cannot be linked) and inference (personal information cannot be deduced). The techniques behind the steps are described on the page about PII removal and de-identification before licensing.

Data the Commission Considers Hard to Anonymize

The Commission d'accès à l'information states that it is almost impossible to certify that anonymized information could never be re-identified, and that some information is so distinctive that it cannot be adequately anonymized, naming genetic, biometric and geolocation data.

On that view, voice, face, location and DNA datasets offered for training will in most cases remain personal information under Quebec law, with the consent and section 17 duties attached.

Showing Data to a Prospective Buyer: The Commercial-Transaction Exception (Section 18.4)

Since September 22, 2022, section 18.4 lets a business communicate personal information without consent to the other party of a commercial transaction it intends to enter, where the communication is necessary to conclude that transaction.

A commercial transaction is the sale or lease of all or part of a business or its assets, a merger or other change of legal structure, a loan or other financing, or security for an obligation.

A prior agreement must bind the other party to use the information only to conclude the transaction, not to pass it on without consent unless the Act allows, to protect its confidentiality, and to destroy it if the deal does not close.

After closing, the other party may use or communicate the information only under the Act and must notify the person within a reasonable time that it holds the information because of the transaction.

Whether a license of a dataset whose value is the personal information itself counts as a "commercial transaction" is not decided by any source read for this page, so the safe reading for a sample is the narrow one.

Describing a dataset without handing it over avoids it, as the data manifest that describes a dataset without handing it over shows.

Sending Personal Information Outside Quebec: The Section 17 Assessment and Written Agreement

Before a business communicates personal information outside Quebec, section 17 requires a privacy impact assessment, so a training set sent to a buyer in Ontario, the United States or anywhere else needs one first. The assessment must consider, among other things:

  • the sensitivity of the information;
  • the purpose of its use;
  • the protection measures, including contractual ones, it would receive;
  • the legal regime of the place it would go, including its privacy principles.

The communication may proceed if the assessment shows adequate protection, in particular in light of generally recognized privacy principles, and it must be covered by a written agreement that takes account of the assessment's results and any agreed risk-mitigation terms.

The same applies when a business entrusts a person outside Quebec with collecting, using, communicating or keeping information (s.17, in force 2023-09-22), and on the text's wording another Canadian province counts as outside Quebec. The written agreement is a clause set in the license, covered under AI data licensing agreements.

A Privacy Impact Assessment for New Information Systems (Section 3.3)

Section 3.3 adds a second assessment duty on the buyer side: any project to acquire, develop or overhaul an information system or electronic service delivery that involves personal information needs a privacy impact assessment, with the person in charge of privacy protection consulted from the start.

The assessment is proportionate to the sensitivity, purpose, quantity, distribution and medium of the information. No ruling decides whether a given AI model project is an "information system".

Penalties, Offences and the Commission d'accès à l'information

The Commission d'accès à l'information can impose administrative monetary penalties of up to CAD 10,000,000 or 2% of worldwide turnover of the preceding fiscal year, whichever is greater, and penal fines for businesses run from CAD 15,000 to CAD 25,000,000 or 4% of worldwide turnover, whichever is greater, as of 2026-11-17. For natural persons the administrative maximum is CAD 50,000 and the penal range is CAD 5,000 to CAD 100,000.

Offences include handling personal information contrary to the Act and identifying or trying to identify a person from de-identified information without authorization or from anonymized information.

The Commission was one of the four offices in the 2026 joint investigation into OpenAI's training data, which published its findings on 2026-05-06. The conclusions of each office are set out in the Canadian privacy commissioners' OpenAI investigation.

Benefits, Risks, Downsides and Alternatives for a Seller Under Law 25, Including Not Selling

Law 25 lets a Quebec business license data for AI training, but personal information brings purpose-specific consent and a section 17 assessment with it, and licensing only anonymized data or not licensing at all remain valid choices.

Benefits

  • Information anonymized under the regulation leaves the Act.
  • Section 18.4 allows a controlled look by a prospective buyer under agreement.
  • A documented consent per purpose is portable proof for the buyer.

Risks

  • Penal fines reach CAD 25,000,000 or 4% of worldwide turnover (as of 2026-11-17).
  • Biometric, genetic and geolocation sets will rarely count as anonymized.
  • Whether training is a compatible purpose remains open.

Downsides

  • A section 17 assessment and a written agreement come before any transfer outside Quebec.
  • The regulation requires risk analyses, periodic re-assessment and a register.
  • Consent is requested again for each new purpose.

Alternatives

  • License only anonymized data.
  • License data that holds no personal information.
  • Keep the data and use it internally within the original purposes.
  • Not selling or licensing at all.

The full range of alternatives to selling data to AI labs is covered on its own page.

Beyond Quebec: Federal Rules, Other Privacy Regimes and the Seller Steps

Quebec's Act sits beside federal law. The Canada rules for licensing data for AI training page covers PIPEDA, the pending Bill C-36, the Copyright Act and withholding tax, while the pages below compare other regimes and apply the rules to a sale.

How This Page Is Sourced and Dated

Every rule here cites the section of Quebec's private-sector Act, the assented text of Law 25, the Gazette officielle or the Commission's own pages, read in French and checked on November 17, 2026.

The methodology page explains how we source, date and check every figure. This page is general information about Quebec law, not legal advice for any specific deal.

Frequently Asked Questions

Does Quebec Law 25 apply to AI training data?

Yes, whenever a training set holds personal information a business in Quebec collected. A new training use needs consent unless a section 12 exception applies, and anonymized data falls outside the Act.

Not as a general rule, because de-identified data is still personal information. Only the study, research or statistics exception may apply, and whether training qualifies is unsettled.

Is a privacy impact assessment needed to send training data to a buyer outside Quebec?

Yes. Section 17 requires a privacy impact assessment and a written agreement before personal information is communicated outside Quebec, in force since September 22, 2023.

What are the fines under Quebec Law 25?

Administrative penalties reach CAD 10,000,000 or 2% of worldwide turnover, and penal fines reach CAD 25,000,000 or 4%. In each case the greater amount applies, as of 2026-11-17.