ISO 42001 for Canadian organisations, from an AI system inventory to a certificate an accredited body signs.
One Canadian team inventories the AI, builds the AI management system, and prepares both audit stages.
Annex A carries 38 controls. A Statement of Applicability decides which of them bind you, and every control you leave out has to be justified in writing.
Policy, roles, impact assessment, lifecycle records, data provenance and supplier terms, written so an auditor can follow the AIMS end to end rather than take your word for it.
Stage 1 reads your documents. Stage 2 tests whether what you documented actually runs, and the gap between the two can't exceed six months or stage 1 is repeated in full.
ISO 42001 certification isn't theoretical any more. Over 350 organisations worldwide held the certificate by April 2026, and Canada has exactly one accreditation body offering it.
IN PRODUCTION
A European tender asked for ISO 42001 and we had eleven weeks. Two consultancies quoted us a year and a policy library. SMEnode started by listing what we actually run, which turned out to be four systems, not the fourteen we'd assumed. Smaller scope, real evidence, certificate in hand before the deadline.
No Canadian law requires this. A customer, a tender, or a board asks for it, and that turns out to be a firmer deadline.
Before anything else we write down every AI system you run, who owns it, what it decides, whose data feeds it, and which of them a supplier operates on your behalf. That list is usually shorter than the internal assumption and it always contains one surprise: a model somebody bought on a credit card, or a vendor feature switched on by default. Scope comes from that list. Annex A's 38 controls then get filtered through a Statement of Applicability, and you justify every one you leave out. Then comes the unglamorous part. An impact assessment saying what happens to a person when the system is wrong. Lifecycle records showing what changed and who approved it. Data provenance you can show. Supplier terms that survive the question of who trained this. Most gaps we find aren't missing controls, they're controls with nothing written down, and stage 1 reads documents before stage 2 tests anything.
ISO/IEC 42001:2023 is the first international AI standard written for the management system around your models rather than for the models themselves. Buyers often ask for it as responsible AI certification, which is close enough as shorthand but not what the certificate says. The next question is always ISO 42001 vs ISO 27001. The answer is subject, not structure: both are management system standards with the same clause layout, so holding one makes the second materially cheaper. ISO 27001 governs information security, who can reach data and how it's protected. ISO 42001 governs AI systems, what a model decides, whose data trained it, and what happens when it's wrong. Same auditors, same shape, different risk.
Canada still has no AI statute in force, so nothing here is legally compulsory. AIDA halted in January 2025 and no successor has been tabled, which leaves PIPEDA, Quebec Law 25 and the voluntary code as what actually applies. The EU AI Act's high-risk deadline, the one most of the market sells against, moved from 2026-08-02 to 2027-12-02 when the Digital Omnibus on AI, Regulation (EU) 2026/1744, took effect on 2026-07-27. Article 50 transparency duties were excluded from that deferral and landed on time. So the honest reason to certify is commercial. Somebody won't buy without it, and that's a better deadline than a regulatory one because it doesn't move.
Four steps, and step 01 is counting things. Nobody wants to start with an inventory and it decides the cost of everything after it. A team that thinks it runs fourteen AI systems often runs four that matter, plus ten features inside software it already pays for. Certifying the four is months of work. Certifying fourteen is a year you needn't spend. We won't quote the rest until this is done.
Every system, owner, decision, data source and supplier written down, then a defensible boundary drawn around the ones that belong inside the management system. The surprise in that list is why we do it first.
The 38 Annex A controls filtered against your risk and impact assessments, each exclusion justified in writing, because stage 1 will ask about the exclusions and not the inclusions.
Policy, roles, lifecycle records, data provenance and supplier terms. Our engineers do the technical work rather than handing you a folder of templates to fill in yourselves.
We prepare you for stage 1, fix what it raises, and stand behind you through stage 2. The certificate comes from an accredited body, never from us, and we'll help you pick one whose accreditation is real.
Answers first, including whether this standard is mature enough to bother with and why the EU deadline moved. An architect takes the call, not a salesperson.
LET'S CONNECT
A senior engineer replies within an hour, 24/7.