ISO/IEC 42001 certificates for AI management systems now appear in vendor questionnaires, sales decks and trust centres, but a certificate is only as credible as the body that issued it and the audit behind it. ISO/IEC 42006:2025 sets the requirements for bodies that audit and certify AI management systems (AIMS): auditor competence, how much audit time is needed, what the certificate must state and how certification is maintained. This guide explains the standard from the buyer's side: what it requires, how to tell an accredited, well-scoped 42001 certificate from a weak one, and what to ask a supplier whose AI system will hold credentials or act on your data. It is written for procurement, third-party risk, AI governance and security teams, and for organisations choosing their own certification body. It complements the ISO/IEC 42001 Guide, which covers the management system itself, and describes the standard from its catalogue entry, published table of contents and accreditation body notices without reproducing its text.
Key takeaways
- ISO/IEC 42006:2025 governs the certifier, not the certified. Edition 1, published July 2025 by ISO/IEC JTC 1/SC 42, it adds AI-specific requirements to ISO/IEC 17021-1 for bodies certifying against ISO/IEC 42001.
- Accreditation is the first test. A 42001 certificate from a body accredited for AIMS by a recognised accreditation body carries weight; one from an unaccredited issuer does not, however professional it looks.
- Scope decides what the certificate covers. The standard has a dedicated clause on AIMS certification documents; read the scope to see which AI systems, roles and sites are in, and which are not.
- Audit time and auditor competence now have rules. Clause 9.1.4 and normative Annex A set how audit time is determined, and clause 7.1 sets generic and specific technical competence requirements for personnel.
- A certificate is not evidence about your use case. It shows the supplier runs an AI management system; it does not show how their agents authenticate, what credentials they hold or how they would handle your data. Ask those questions separately.
What it is and who it applies to
| Item | Detail (ISO catalogue, checked 8 November 2026) |
|---|---|
| Reference | ISO/IEC 42006:2025 |
| Title | Information technology, Artificial intelligence, Requirements for bodies providing audit and certification of artificial intelligence management systems |
| Edition and date | Edition 1, July 2025 |
| Status | Published, stage 60.60 |
| Length | 31 pages |
| Committee | ISO/IEC JTC 1/SC 42 (Artificial intelligence) |
| Builds on | ISO/IEC 17021-1 (requirements for bodies certifying management systems); can also be integrated into ISO/IEC 17065 schemes for AI products, processes or services |
| Intended users | Certification bodies auditing AIMS against ISO/IEC 42001, and accreditation bodies assessing them |
- Why buyers should care: before 42006, AIMS certification bodies worked from the general 17021-1 rules plus draft criteria. The Standards Council of Canada, for example, states that it used the DIS or FDIS as temporary accreditation criteria before publication and moved accredited bodies to the published standard with immediate effect.
- What changed between draft and final: the same SCC bulletin lists revisions after the DIS to liability (5.3.2), audit time calculation (9.1.4.2 and Annex A), competence for certification personnel (7.1 and 7.2), AIMS certification documents (8.2.2) and access to the organisation's documentation (8.4.2), with only editorial changes after the FDIS.
- The parallel for security: the equivalent standard for ISO/IEC 27001 certification bodies is ISO/IEC 27006-1. Buyers who already check 27001 certificates can apply the same discipline. See the ISO 27001 Certification Body Selection guide.
Structure and key requirements
The published table of contents follows the 17021-1 structure. The table maps each clause to what it means for someone relying on a certificate. The interpretations are NHI Mgmt Group's, not text from the standard.
| Clause | Topic | Why it matters to a buyer |
|---|---|---|
| 5.2 | Management of impartiality, conflicts of interest | A body that also advised the supplier on implementing its AIMS should not certify it |
| 5.3 | Liability and financing | Revised after the draft; the body must be able to cover liabilities from its certification activities |
| 7.1 | Competence of personnel: generic and specific technical competence | Auditors need AI-specific knowledge, not only management system audit skills |
| 7.2 | Personnel in certification activities, demonstration of knowledge and experience | Competence must be demonstrated and recorded, not assumed |
| 7.3, 7.5 | External auditors, technical experts, outsourcing | Specialist experts can join audits; outsourcing is controlled |
| 8.2.2 | AIMS certification documents (Annex C gives a template) | Defines what a 42001 certificate should state; compare the certificate you receive against it |
| 8.4.2 | Access to the organisation's documentation | Addresses auditors seeing sensitive material such as model and data documentation |
| 9.1.3 | Scope of certification | The scope must be defined; vague scopes are a warning sign |
| 9.1.4 and Annex A | Determining audit time (normative), with examples in Annex B | Prevents token audits; a very short audit for a complex AI estate should raise questions |
| 9.1.5, 9.1.6 | Multi-site sampling, multiple management systems | Integrated 27001 and 42001 audits are possible but must still cover AI-specific requirements |
| 9.2.4 | Deployment of remote audit | Remote audits are permitted under defined conditions |
| 9.6 | Maintaining certification: surveillance, re-certification, special audits, suspension and withdrawal | A certificate's status can change; check it is current |
Credible and weak certificates compared
| Signal | Credible | Weak or misleading |
|---|---|---|
| Issuer | Certification body accredited for ISO/IEC 42001 by a national accreditation body that is a signatory to the International Accreditation Forum (IAF) multilateral arrangement | Unaccredited issuer, or accreditation from a body outside recognised arrangements |
| Standard | ISO/IEC 42001:2023 named on the certificate | "Compliant with" or "aligned to" 42001, a self-declaration, or a training certificate presented as certification |
| Scope | Names the AI systems, products or services, the organisation's role (for example provider or user of AI) and sites | Generic scope such as "AI activities" or a scope limited to one small unit while marketing implies the whole company |
| Validity | Issue and expiry dates; status verifiable with the issuer or a public register | No expiry, no way to verify, or status not checked since issue |
| Supporting evidence | Supplier will share Statement of Applicability or an audit summary under NDA | Supplier refuses any detail beyond the certificate image |
What ISO/IEC 42006 does not do
- It does not certify anyone directly. It is used by accreditation bodies to assess certification bodies, which then certify organisations against ISO/IEC 42001.
- It does not add requirements for the certified organisation. The AIMS requirements stay in ISO/IEC 42001; 42006 governs how conformity with them is audited.
- It does not replace ISO/IEC 17021-1. The general rules for management system certification bodies, ISO/IEC 17021-1:2015 (confirmed 2020), still apply; 42006 adds AI-specific requirements on top.
- It does not judge an AI product's safety or accuracy. Product, process or service conformity assessment is a separate route, which the catalogue notes can use ISO/IEC 17065 schemes.
Identity and NHI requirements
Neither ISO/IEC 42006 nor ISO/IEC 42001 sets technical identity controls for AI systems. That is a gap buyers must fill, because AI services and agents increasingly hold OAuth tokens, API keys and delegated permissions into the buyer's systems. A 42001 certificate tells you a governance system exists; it does not tell you how these credentials are issued, scoped, stored, rotated or revoked.
- Ask how the supplier's AI agents authenticate: per-agent identities with short-lived credentials, or shared static keys. See the Agent Identity and Access Guide.
- Ask what access the integration needs into your estate: OAuth scopes, service accounts or API keys, and whether you can restrict and revoke them yourself.
- Check whether identity is inside the certified scope: if the AI product that will connect to your systems is outside the 42001 scope, the certificate says nothing about it.
- Pair the certificate with security evidence: an ISO/IEC 27001 certificate whose scope covers the same service, or an independent assurance report, gives more relevant evidence about credential handling than 42001 alone.
- Use a structured questionnaire: the AI Vendor Security Questionnaire Template includes identity and data handling questions that a certificate does not answer.
Implementation steps
For buyers, implementation means building certificate checks into AI procurement and third-party risk. For organisations seeking their own 42001 certificate, the same steps apply to selecting a certification body.
- Add 42001 evidence requests to AI intake. When a supplier claims ISO/IEC 42001 certification, request the certificate, its scope and the issuing body's name. Record it in the supplier record.
- Verify accreditation. Check that the issuing body is accredited for ISO/IEC 42001 by its national accreditation body, and confirm the certificate on the issuer's register or IAF CertSearch where available.
- Read the scope against your use case. Confirm that the AI system, product or service you are buying, and the supplier's role in providing it, are inside the certified scope.
- Check currency. Confirm the certificate has not expired, been suspended or been reduced in scope. Diary a re-check at the next surveillance point, typically annually under 17021-1 certification cycles.
- Ask for supporting detail. Request the Statement of Applicability and, where offered, an audit summary or management letter under NDA. Note material exclusions.
- Close the identity and security gaps. Apply your AI vendor questionnaire, focusing on credentials, delegated access, data use and incident notification. See the AI Procurement Checklist.
- Decide and record. Weight the certificate as one input to the risk decision, with explicit conditions (for example, scope expansion by renewal, or contractual commitments on credential handling).
- If choosing your own certification body: confirm AIMS accreditation, ask how audit time will be calculated under Annex A, request the audit team's AI competence profiles, and confirm the body did not provide implementation consultancy to you.
Decision guidance
- Low-risk AI tools: an accredited 42001 certificate with a relevant scope may be enough with a light questionnaire.
- High-risk or agentic integrations: treat the certificate as a governance indicator only. Require security evidence about credentials and access, contractual controls and the right to audit.
- Startups without certification: absence of a 42001 certificate is not a failure in itself; certification is voluntary. Assess the controls directly with a structured assessment. See the AI Risk Assessment Guide.
- Integrated audits: a combined 27001 and 42001 audit by one body is efficient, but ask whether the audit time and team covered AI-specific requirements, as clause 9.1.6 on multiple management systems contemplates.
Evidence and assessment
| Evidence to collect from a supplier | What to check |
|---|---|
| ISO/IEC 42001 certificate | Issuer, accreditation mark, scope, sites, issue and expiry dates |
| Accreditation confirmation | Issuing body accredited for AIMS; accreditation body recognised |
| Register or verification result | Certificate status current, not suspended or withdrawn |
| Statement of Applicability | Annex A controls of ISO/IEC 42001 applied or excluded, with justification |
| Audit summary or management letter (if offered) | Open nonconformities, areas of concern |
| Identity and security questionnaire responses | Agent authentication, credential scope and lifetime, revocation, logging |
Metrics
- Verification rate: percentage of supplier 42001 claims verified against an accredited issuer's register.
- Scope match rate: percentage of certified suppliers whose certified scope covers the product you buy.
- Currency: percentage of supplier certificates re-checked within the last 12 months.
- Identity coverage: percentage of AI suppliers with access to your systems who have answered credential and delegated access questions.
- Rejected claims: number of unaccredited or misdescribed certificates identified, a sign that checks are working.
Mapping to other frameworks
CSF 2.0 and SP 800-53 identifiers were checked against the official NIST datasets.
| Buyer activity | ISO/IEC 27001:2022 Annex A | NIST CSF 2.0 | NIST SP 800-53 Rev. 5 |
|---|---|---|---|
| Know and tier AI suppliers | 5.19 Information security in supplier relationships | GV.SC-04 | SR-6 |
| Due diligence before contract, including certificate checks | 5.19; 5.20 Addressing information security within supplier agreements | GV.SC-06 | SA-4, SR-6 |
| Ongoing monitoring of certificate status | 5.22 Monitoring, review and change management of supplier services | GV.SC-07 | SR-6, CA-2 |
| Use of AI cloud services | 5.23 Information security for use of cloud services | GV.SC-07 | SA-4 |
For the management system itself, see the ISO/IEC 42001 Guide and the ISO 42001 Readiness Checklist; for auditor training routes, see the ISO 42001 Lead Auditor and Lead Implementer Guide. ISO/IEC 42005 on AI system impact assessment is covered in the ISO/IEC 42005 AI System Impact Assessment Guide.
Common pitfalls
- Accepting the certificate image. A PDF logo is not verification; certificates can be expired, withdrawn or issued by unaccredited bodies.
- Confusing training with certification. An individual's 42001 lead auditor or implementer course certificate says nothing about the organisation's AIMS.
- Ignoring scope. A certificate covering one internal AI programme is used to imply that every product is certified.
- Treating 42001 as a security attestation. It certifies an AI management system; it does not test how agents, models or integrations protect credentials and data.
- Not checking impartiality. When the certification body also sold the implementation consultancy, the audit's independence is in doubt.
- Checking once. Certificates are maintained through surveillance and can be suspended; supplier records go stale without a re-check date.
Practitioner checklist
- Request the full 42001 certificate, scope and issuer name for every AI supplier claiming certification.
- Verify the issuer's AIMS accreditation and confirm the certificate is current on a register.
- Confirm the product or service you are buying is inside the certified scope.
- Ask for the Statement of Applicability and note excluded controls.
- Send an identity questionnaire covering agent authentication, credential scope, lifetime and revocation.
- Pair 42001 with security assurance evidence covering the same service.
- Set a re-check date for certificate status at least annually.
- When selecting your own certification body, ask for audit time calculation under Annex A and team AI competence.
Standards and references
- ISO: ISO/IEC 42006:2025 Requirements for bodies providing audit and certification of artificial intelligence management systems
- ISO: ISO/IEC 17021-1:2015 Requirements for bodies providing audit and certification of management systems
- Standards Council of Canada: Transition to ISO/IEC 42006:2025
- DIN: ISO/IEC 42006:2025 table of contents
- NIST: Cybersecurity Framework 2.0
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: ISO/IEC 42001 Guide · ISO 27001 Certification Body Selection · AI Vendor Security Questionnaire Template · Agent Identity and Access Guide