ISO/IEC 42001 certification: the AI management system standard, and what it involves
ISO/IEC 42001 is the international standard for an artificial intelligence management system, and certification against it is granted by an accredited third-party certification body after an audit, not by ISO. This page defines the term, sets out what an auditor is looking at when they assess you, and points to the neighbouring terms compliance officers meet in the same conversation.
Auditrek is not open yet, so anything on this page about what it does is what it is being built to do.
What ISO/IEC 42001 is
ISO/IEC 42001 is the international standard that specifies requirements for an artificial intelligence management system, meaning the policies, assigned roles, risk assessments, oversight arrangements and records an organisation maintains around the AI systems it develops or deploys.
It is a management system standard, which puts it in the same family as ISO/IEC 27001 for information security and ISO 9001 for quality. That family resemblance matters more than it first appears. A management system standard certifies how an organisation governs an activity over time; it does not certify that any individual model is accurate, fair or safe. An auditor assessing you against ISO/IEC 42001 is asking whether you know which AI systems you run, whether you assessed their risks, whether someone owns each one, whether human oversight exists where you said it does, and whether you can show that those arrangements actually operated rather than merely being written down.
An example
A mid-market insurer runs a claims triage model that routes a subset of claims to manual review. The policy binder says an adjuster reviews every automated decline. At the audit, the policy binder is the easy part. The harder question is evidential: for the review period, which decisions did the model make, which ones were overridden by a human, and can the record be shown to be unaltered since it was written. Organisations that treat governance as a document exercise tend to discover that gap late, during the stage where operating effectiveness is being tested rather than described.
What certification generally involves
- Define the scope
Decide which AI systems, business units and jurisdictions sit inside the management system. Scope decisions made loosely here tend to reappear as audit findings later, because everything in scope needs evidence behind it.
- Run a gap assessment
Compare the standard's requirements against what your organisation already does. Most regulated enterprises already hold parts of this from model risk management or information security work, and the gap is usually in records rather than in policy.
- Build and operate the management system
Write the policies, assign ownership, run the risk assessments, and then let the system run for long enough to produce evidence. An auditor tests operation, so the management system needs a history, not just a launch date.
- Stage one: documentation review
The certification body reviews your scope, policies and readiness, and reports where the system is not yet auditable. This stage is largely a reading exercise against what you have written.
- Stage two: audit of operating effectiveness
The auditor samples. They pick systems, periods and decisions and ask for the underlying records. This is the stage where retrievability of evidence, rather than its existence somewhere in a data lake, determines how the week goes.
- Surveillance and recertification
Certification under an ISO management system standard is maintained through periodic surveillance audits rather than granted once. Confirm the cycle and its timing with your chosen certification body, since it is set by them and by their accreditation rules.
Related terms
- AI management system (AIMS): the governed set of policies, roles, controls and records that ISO/IEC 42001 describes requirements for. The certificate attaches to the management system, not to a model.
- Certification body and accreditation body: the certification body audits you and issues the certificate; the accreditation body oversees certification bodies. ISO publishes the standard and does not certify organisations against it.
- Gap assessment: the pre-audit comparison of current practice against the standard's requirements, usually the first paid engagement in a certification programme.
- Conformity assessment under the EU AI Act: a separate, statutory process for high-risk AI systems placed on the EU market. It runs on regulation, not on a voluntary standard, and an ISO/IEC 42001 certificate is not a substitute for it, though the underlying evidence often overlaps.
- Record-keeping and logging obligations, Article 12: the EU AI Act's requirement around automatic logging over an AI system's lifetime. Auditors and supervisory authorities are both asking about records, which is why one evidence base tends to serve both.
- Tamper-evident record: a record whose alteration is detectable, typically through a hash chain where each entry incorporates the hash of the one before it. Distinct from an immutable store, which prevents alteration at the storage layer.
- NIST AI Risk Management Framework: the voluntary US framework often run alongside or ahead of ISO/IEC 42001 work. Our guide on turning its Measure and Manage functions into retrievable evidence covers how the two lines of work share an evidence base.
Questions
Who issues ISO 42001 certification?
An accredited third-party certification body, not ISO. ISO develops and publishes the standard; certification bodies audit organisations against it and issue certificates, under the oversight of a national or regional accreditation body.
Does ISO 42001 certification satisfy the EU AI Act?
No. The EU AI Act is regulation with its own conformity assessment and record-keeping obligations, and a voluntary standard certificate is not a legal substitute for them. The practical overlap is in evidence: the AI inventory, risk assessments, human oversight arrangements and decision records that an auditor samples are largely the same material a supervisory authority would ask about. Treat the legal question as one for qualified regulatory counsel.
What evidence do ISO 42001 auditors actually ask for?
Expect sampling. The auditor picks systems and periods and asks for the records behind them: what the system decided, when, on which model version, whether a human overrode it, and who owned the system at the time. The difficulty for most organisations is not having the data at all, it is producing a specific period's records quickly and showing they have not been edited since.
Is ISO 42001 the same kind of certificate as ISO 27001?
Structurally, yes. Both are management system standards built on the same high-level structure, both are audited in stages and maintained through surveillance, and both certify governance of an activity rather than the quality of any single output. If your organisation already holds ISO/IEC 27001, the programme shape will be familiar and some controls will carry over.
How long does certification take?
It depends on scope and on how much evidence already exists, and your certification body sets the audit timetable. The constraint most teams underestimate is that operating effectiveness is tested over a period, so the management system needs to have been running and producing records before stage two, not assembled in the weeks before it.
Does Auditrek certify us against ISO 42001?
No. Auditrek is not a certification body, does not assess organisations against the standard, and its reports are structural records of what an AI system did and when. Certification decisions sit with an accredited certification body, and interpretation of your obligations sits with your counsel.