Canada has no artificial intelligence law. The Artificial Intelligence and Data Act (AIDA) died on the order paper when Parliament was prorogued in January 2025. Many executives concluded they had time.
Our position: your European customer will apply the AI Act before your government does, and it will apply it to you by contract. The European regulation does not stop at the Union's borders: it follows the product and its output. It is a position, not a certainty: many Canadian companies have no AI use that falls within the text, and Europe itself has just postponed part of its obligations. But those selling software, equipment or a service with AI inside in Europe will get the questions long before they have a Canadian law to read.
A regulation that follows the product, not the company
Regulation (EU) 2024/1689, known as the AI Act, entered into force on 1 August 2024. Its Article 2 sets the scope, and it is the part Canadian companies read least. It covers, among others:
- providers placing an AI system on the Union market or putting it into service, whether they are established in the Union or in a third country;
- providers and deployers established outside the Union where the output of the system is used in the Union;
- importers, distributors and product manufacturers placing an AI system on the market under their own name.
Status: confirmed, text of the regulation.
In practice, a Quebec HR software vendor selling to a French subsidiary, a manufacturer whose equipment embeds a vision module sold in Germany, or a services firm delivering model-produced analyses to a Belgian client are all in scope. The first is probably a provider. The second makes a product that integrates AI. The third may be a deployer whose output is used in the Union.
The timeline, after the Digital Omnibus
The text applies in stages. Here is the state as of 15 October 2026, according to the European Commission:
- 2 February 2025: prohibited practices (harmful manipulation, social scoring, some biometric identification, etc.) and the AI literacy obligation of Article 4: providers and deployers must ensure their staff have a sufficient level of AI literacy. Status: confirmed.
- 2 August 2025: obligations for providers of general-purpose AI models (technical documentation, copyright policy, training data summary) and governance rules, with the AI Office. Status: confirmed.
- 2 August 2026: the regulation becomes broadly applicable, including the transparency rules of Article 50: telling people they are dealing with an AI, making generated content identifiable, labelling deepfakes. Status: confirmed.
- 2 December 2026: a ninth prohibited practice, added by the Omnibus, applies to systems generating non-consensual intimate imagery or child sexual abuse material. Generative AI providers already on the market before 2 August 2026 have until that date for the content marking of Article 50(2), according to several law firms. Status: confirmed by the Commission for the prohibition; reported for the marking grace period.
- 2 December 2027: obligations for high-risk systems in Annex III (recruitment, credit, education, access to essential services, etc.), postponed from August 2026. Status: confirmed.
- 2 August 2028: obligations for high-risk systems embedded in regulated products (Annex I: machinery, medical devices, toys, etc.). Status: confirmed.
The postponement comes from the Digital Omnibus on AI, proposed by the Commission in November 2025. Provisional trilogue agreement on 7 May 2026, European Parliament vote on 16 June, Council adoption on 29 June, publication in the Official Journal on 24 July as Regulation (EU) 2026/1744, entry into force on 27 July, six days before the original deadline. Status: dates reported by several law firms (Bird & Bird, Gibson Dunn); new deadlines confirmed by the Commission.
What it changes: the postponement moves the date, not the architecture. Risk tiers, roles and obligations stay the same. A company that waits until December 2027 to start will find that its European customers are not waiting: they are already preparing their own deployer obligations and passing them on to their suppliers.
Four risk tiers, and one useful question
The regulation sorts AI systems into four tiers:
- Unacceptable risk: prohibited.
- High risk: risk management system, quality data, technical documentation, logging, human oversight, robustness and cybersecurity, conformity assessment, registration in the EU database.
- Transparency risk: inform, mark, label.
- Minimal risk: no obligation specific to the regulation, beyond AI literacy.
Status: confirmed by the European Commission.
For an executive team, the only useful starting question is: which of our AI uses touches a decision about a person, or a regulated product sold in Europe? An assistant summarizing internal meeting notes is almost always minimal risk. A module that screens job applications, assesses creditworthiness or controls a safety function of a machine is probably high risk.
Penalties: comparable to the GDPR, and beyond
Article 99 provides for administrative fines of up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher, for prohibited practices. Other breaches can reach EUR 15 million or 3%, and incorrect information supplied to authorities EUR 7.5 million or 1%. For SMEs, the lower of the two amounts applies. Status: confirmed, text of the regulation.
What it changes: for a Canadian SME, the fine is not the first risk. The first risk is commercial: a European customer who cannot demonstrate its own deployer compliance will not renew a supplier who does not give it the documentation it needs.
What a Canadian company selling in Europe should do
Five steps, in this order.
- Inventory. List the AI systems embedded in what you sell in Europe, and those whose output is used there. Internal tools come next. Without an inventory, none of the next steps is possible; it is also the first question in any customer questionnaire.
- Classify. For each system, the risk tier, with a one-line justification. Most will land in minimal risk or transparency. What matters is being able to show the question was asked.
- Qualify your role. Provider, deployer, importer, product manufacturer: obligations change completely. A vendor that integrates a third-party large model into its product and sells it under its own name is the provider of the system, even if it did not train the model.
- Document. For high-risk systems, the Annex IV technical documentation: purpose, data, performance, limits, human oversight measures, logging. Start now: it is the longest deliverable, and the one your customer will ask for first.
- Train. The AI literacy obligation of Article 4 has applied since February 2025 to every provider and deployer in scope, whatever the risk tier. Short, documented, role-based training is enough to start.
ISO/IEC 42001: the bridge, not the pass
ISO/IEC 42001, published in December 2023, describes an AI management system: policy, roles, risk and impact assessment, life cycle, suppliers. It does not replace a conformity assessment under the AI Act, and it is not, to date, a harmonized standard under the regulation. Status: confirmed by ISO for publication; European harmonized standards are still being prepared at CEN-CENELEC.
It remains the best bridge for a Canadian company: a single management system, recognized on both sides of the Atlantic, that produces the inventory, classification and documentation the AI Act requires. We explained it in ISO 42001 or NIST AI RMF: where to start.
And Canada, meanwhile?
Without AIDA, there is no federal AI-specific law. But no law does not mean no obligations: in Quebec, Law 25 already governs decisions based exclusively on automated processing and privacy impact assessments. Sector regulators are moving too, such as OSFI with its Guideline E-23 on model risk management. A program built for the AI Act covers a large share of what these texts require.
What we recommend to an executive team
The AI Act is not a reason to slow down on AI. It is a format for evidence. Canadian companies that adopt it early will sell more easily in Europe, and answer faster the day Canada legislates. Those that wait will receive their customers' format, with their customers' deadlines.
We think an SME selling in Europe should have, by the end of the year, an inventory, a classification and a named owner. Too early for a 50-person company? Or already late for those selling to European banks and manufacturers? Tell us in the comments where you stand.
FortaRisks' Compliance module maps the AI Act, ISO/IEC 42001 and more than 40 frameworks to your existing controls, so you document once. Our free AI governance assessment places you in five minutes, and thirty minutes with an expert is enough to classify your first three uses.
Sources: Regulation (EU) 2024/1689 on artificial intelligence, Official Journal of the European Union · Regulation (EU) 2026/1744, Digital Omnibus on AI · European Commission, regulatory framework on AI · ISO, ISO/IEC 42001:2023. Monitoring, named in the text: Bird & Bird, Gibson Dunn.