Standards and law: 42001, 42005, 42006 and the EU AI Act
Four documents, one job each
The same harmonised structure as 27001: context and AIMS scope (4), leadership and AI policy (5), AI risk assessment, AI system impact assessment and objectives (6), resources and competence (7), operation and life-cycle controls (8), performance evaluation (9), improvement (10).
A.2 AI policies, A.3 internal organisation, A.4 resources for AI systems, A.5 assessing impacts of AI systems, A.6 AI system life cycle, A.7 data for AI systems, A.8 information for interested parties, A.9 use of AI systems, A.10 third-party and customer relationships. Annex B gives implementation guidance per control, Annex C organisational objectives and risk sources, Annex D use across domains.
Published 28 May 2025; guidance on when and at which life-cycle stages to assess impact, how to document it and how to integrate it into risk management and the management system — supporting 42001 clauses 6.1.4 and A.5, and usable as the method for the EU AI Act Article 27 fundamental rights impact assessment.
Published 7 July 2025; adds AIMS-specific competence and consistency requirements on top of ISO/IEC 17021-1. When choosing a certification body, confirm it is accredited under ISO/IEC 17021-1 and ISO/IEC 42006 so the certificate is comparable.
Annex A maps closely to AI Act Articles 9 (risk management), 10 (data governance), 11–12 (technical documentation, records), 13 (transparency), 14 (human oversight), 15 (accuracy and cybersecurity) and 17 (quality management system) — the backbone for meeting the obligations. But conformity assessment, EU database registration, the authorised representative and Article 50 transparency are statutory procedures a certificate cannot replace. On 14 Oct 2026 the first AI Act harmonised standard was cited in the Official Journal: EN 18286:2026 (Article 17 quality management system; Commission Implementing Decision (EU) 2026/2274). Applying it gives presumption of conformity for the Article 17 requirements it covers, and it is explicitly designed to layer on ISO/IEC 42001 or ISO 9001; a 42001 certificate alone still gives no presumption, and harmonised standards for the other articles are not yet cited.
Annex A controls × EU AI Act articles (extract)
An extract of the mapping we deliver: left, the 42001 control; middle, the corresponding AI Act article; right, how one piece of evidence satisfies both.
| ISO/IEC 42001 Annex A | EU AI Act | One piece of evidence |
|---|---|---|
| A.2 AI policy | Article 17 quality management system (policies and procedures) | One approved AI policy stating the regulatory compliance commitment |
| A.5 AI system impact assessment (+42005) | Article 27 fundamental rights impact assessment; Article 9 risk management | One impact assessment template covering individuals, groups, society and fundamental rights |
| A.6.2 life cycle (requirements, V&V, deployment, monitoring) | Article 9 risk management; Article 15 accuracy, robustness, cybersecurity | Development process documents plus test and validation records |
| A.6.2.7 technical documentation; A.6.2.8 event logs | Article 11 technical documentation (Annex IV); Article 12 record-keeping | Technical documentation index mapped to Annex IV; log retention policy |
| A.7 data (acquisition, quality, provenance, preparation) | Article 10 data and data governance | Datasheets and data quality check records |
| A.8 information for interested parties; A.8.4 incident communication | Article 13 instructions for use; Article 50 transparency; Article 73 serious incident reporting | Instructions-for-use template plus incident reporting flow |
| A.9 use; human oversight | Article 14 human oversight; Article 26 deployer obligations | Intended-use statement and oversight measure design |
| A.10 third parties and customers | Article 25 value-chain responsibilities; Article 26 | Supplier/customer responsibility allocation clauses |
EU AI Act, ISO/IEC 42001 and NIST AI RMF: different things, not substitutes
The most common question: "If we get ISO 42001, are we compliant with the AI Act?" No. The table puts the three side by side; first decide whether your goal is EU market access, a customer-required certificate or internal governance, then choose one or combine them.
| Item | Nature | Who and where | Mandatory? | How it is proven |
|---|---|---|---|---|
| EU AI Act (Regulation (EU) 2024/1689) | EU law (regulation, directly applicable) | AI systems placed on the market or put into service in the EU, and systems whose output is used in the EU — providers, deployers, importers, distributors wherever established (Article 2) | Mandatory; phased (2 Feb 2025 prohibitions/literacy, 2 Aug 2025 GPAI, 2 Aug 2026 Article 50, 2 Dec 2027 Annex III, 2 Aug 2028 Annex I) | High-risk: conformity assessment (mostly internal control, Annex VI), EU declaration of conformity, CE marking, EU database registration; penalties Article 99 |
| ISO/IEC 42001:2023 | International management-system standard (certifiable) | Any organisation, any region or size, providing or using AI; about the organisation's governance and processes, not one product | Voluntary; often required by customer contracts or procurement | Third-party certificate (certification bodies accredited under ISO/IEC 17021-1 and ISO/IEC 42006:2025); 3-year cycle with annual surveillance |
| NIST AI RMF 1.0 | US NIST voluntary framework (Govern / Map / Measure / Manage) | Any organisation; often referenced in US government procurement and by US customers | Voluntary; no certification scheme | Self-declaration or third-party assessment report; no official certificate |
Key differences
- The AI Act governs whether this AI system is lawful in the EU; 42001 governs whether the organisation's AI management is sound. A 42001 certificate does not exempt a high-risk system from conformity assessment, registration, an authorised representative or Article 50 transparency.
- AI Act harmonised standards under Article 40 are developed by CEN-CENELEC JTC 21; the first was cited in the Official Journal on 14 Oct 2026: EN 18286:2026 (quality management system for providers of high-risk AI, Article 17; Commission Implementing Decision (EU) 2026/2274). Applying EN 18286 gives presumption of conformity for the Article 17 requirements it covers; it is designed to be layered on ISO/IEC 42001 or ISO 9001, while a 42001 certificate alone still gives no presumption. Harmonised standards for the other articles are not yet cited.
- The 38 Annex A controls of 42001 map closely to AI Act Articles 9, 10, 12, 14, 15 and 17 — so 42001 is the management backbone for meeting the AI Act, not a replacement.
- The Article 27 fundamental rights impact assessment, 42001 A.5 impact assessment and ISO/IEC 42005:2025 can be satisfied by one assessment.
How to choose
- Product or service enters the EU (including output delivered remotely to EU users) → do AI Act classification and timeline first, with 42001 as the backbone.
- Not entering the EU, but customers (Japanese or European brands, financial institutions) require proof of AI governance → target 42001 certification with an AI Act mapping for later.
- Large internal roll-out of generative AI and worry about control → build use governance from 42001 clauses 6, 8 and A.9; certify later if needed.
What to share when implementing together
When implementing together, use the 42001 management system as the base: AI policy (A.2), system inventory (A.6.2 / AI Act Article 49), risk and impact assessment (6.1 / A.5 / Articles 9, 27), data governance (A.7 / Article 10), technical documentation and logs (A.6.2.7–8 / Articles 11, 12), human oversight (A.9 / Article 14), incident reporting (A.8.4 / Article 73). What remains AI Act-only is statutory procedure: conformity assessment, declaration and CE, database registration, authorised representative.
How Winners works
Winners does all three: AI Act classification and compliance, ISO 42001 implementation and audit support, and integration with ISO 27001/27701. Our first step is always to clarify the goal — which markets, which proof customers want, which risks to control — then recommend one, a sequence or a combination, so you never build two systems for one certificate.
Either entry works: start from your industry's pain points or from the standard you plan to adopt — we first clarify the differences, then give you the information and support.
EU regulatory watch (continuously monitored)
Official sources (European Commission, EUR-Lex, ENISA, CEN-CENELEC, ETSI, EDPB) are monitored continuously; only items relevant to this service are listed. Fed by our on-premise intelligence pipeline; the site refreshes every minute.
Loading…
Method (4 phases)
- Phase 1 | Scope and inventory (2–4 weeks)
Interviews with business, R&D, legal, security; AI system inventory; AIMS scope and roles; integration points with existing 27001/27701.
- Phase 2 | Risk and impact assessment (3–5 weeks)
AI risk assessment per 6.1 and Annex C; impact assessment per 42005 for each in-scope system; AI Act classification in parallel if entering the EU.
- Phase 3 | System build (6–10 weeks)
AI policy, life-cycle and data governance procedures, technical documentation and logging specification, supplier clauses; all 38 Annex A controls implemented with evidence.
- Phase 4 | Audit and certification (4–8 weeks)
Internal audit, management review, corrective action; support through stage 1 and stage 2 (certification body accredited under ISO/IEC 17021-1 and ISO/IEC 42006).
The most cost-effective path to full compliance
The law requires an outcome, not the most expensive method. Below is the lightest lawful path from the official text and our practice, the evidence you can reuse, the steps you can merge, and what must never be cut — cutting it voids the whole compliance.
Lightest lawful path
- Classify before spending: most internal enterprise AI (support chat, summarisation, recommendations) is not high-risk and carries only Article 50 transparency and Article 4 AI literacy duties, not the full high-risk set.[Basis:AI Act Article 6; Annex III; Articles 4, 50]
- Annex III areas limited to narrow procedural tasks, improving completed human results, detecting deviations or preparatory work can be excluded from high-risk under Article 6(3), with a documented and registered assessment.[Basis:Articles 6(3), 6(4), 49(2)]
- Annex III high-risk obligations apply from 2 Dec 2027 and Annex I from 2 Aug 2028; schedule spend accordingly rather than all in 2026, but do Article 50 now since it has applied since 2 Aug 2026.[Basis:Regulation (EU) 2026/1744; Article 113]
- Most high-risk AI can use the internal-control conformity assessment (Annex VI) without a notified body; only Annex III point 1 biometrics needs third parties in specific cases.[Basis:Article 43(1), (2)]
Evidence you can reuse
- An ISO/IEC 42001 management system (risk, data governance, human oversight, records) directly carries Articles 9, 10, 12 and 14; one system serves many AI systems.[Basis:ISO/IEC 42001 clauses 6, 8; AI Act Articles 9–14]
- For products also under the CRA, CRA conformity gives the presumption for Article 15 cybersecurity; no second set.[Basis:CRA Article 12]
- The Article 27 fundamental rights impact assessment and the GDPR Article 35 DPIA can be one assessment (explicitly allowed by Article 27(4)).[Basis:AI Act Article 27(4)]
Steps you can merge
- Merge the AI system inventory with the personal-data RoPA and the CRA product list into one asset register shared by three regimes.[Basis:AI Act Article 49; GDPR Article 30]
- Fold post-market monitoring and serious incident reporting into the existing PSIRT / security incident flow.[Basis:AI Act Articles 72, 73]
Never cut (cutting voids compliance)
- Article 5 prohibited practices cannot be marketed or used since 2 Feb 2025, fines €35M / 7%, with no transition.[Basis:Article 5; Article 99(3)]
- Article 50 transparency (interaction notice, synthetic content marking, deepfake disclosure) is a current obligation since 2 Aug 2026.[Basis:Article 50]
- Non-EU high-risk providers must appoint an authorised representative; high-risk systems must be registered in the EU database.[Basis:Articles 22, 49]
- The Article 6(3) exclusion requires a written assessment and registration; "we are not high-risk" cannot be asserted orally.[Basis:Articles 6(4), 49(2)]
Standards and official sources
- ISO/IEC 42001:2023 — Information technology — Artificial intelligence — Management system (IEC Webstore, 2023-12-18)
- ISO/IEC 42005:2025 — AI system impact assessment (IEC Webstore, 2025-05-28)
- ISO/IEC 42006:2025 — Requirements for bodies providing audit and certification of AI management systems (IEC Webstore, 2025-07-07)
- Regulation (EU) 2024/1689 — Artificial Intelligence Act (EUR-Lex)
- Regulation (EU) 2026/1744 — amending the AI Act application dates (EUR-Lex)
- Commission Implementing Decision (EU) 2026/2274 of 12 October 2026 — harmonised standards for high-risk AI systems (EN 18286:2026), OJ L 14.10.2026
- European Commission — First harmonised standard supporting the AI Act published in the Official Journal (2026-10-19)
- NIST AI Risk Management Framework (AI RMF 1.0)
Last updated 2026-11-07
Learn More About AI Governance
Certification services × risk glossary × latest insights
🏆Certification Services
💡Latest InsightsView all →
No articles yet
Next steps
Book a consultant (free first diagnosis)
60-minute online meeting on your AI system list, customer requirements and current systems: whether to do 42001, the AI Act or both, and the three things to close first.
Book a consultantRequest scope and timeline
Four-phase scope and schedule based on the number of AI systems, existing 27001 and EU market entry.
Request scopeEntering the EU? See the EU AI Act page
Classification engine, application timeline and three deliverable templates.
See EU AI Act service