If your organization already runs an ISO/IEC 27001 information security management system (ISMS), you are much closer to ISO/IEC 42001 certification than you think. Most of the management-system machinery is reusable.…
If your organization already runs an ISO/IEC 27001 information security management system (ISMS), you are much closer to ISO/IEC 42001 certification than you think. Most of the management-system machinery is reusable. The part that isn't is where AI governance actually lives, and it is the part most teams underestimate.
This post walks through both standards side by side: what you can lift straight across from your ISMS, what ISO 42001 adds that has no equivalent in ISO 27001, and how to run both as one integrated system instead of two parallel bureaucracies.
A note on versions: this comparison uses ISO/IEC 27001:2022 and ISO/IEC 42001:2023.
Same Skeleton, Different Subject
Both standards follow ISO's Harmonized Structure, the common template that every modern management system standard is built on. That means the same ten clauses in the same order:
- Scope
- Normative references
- Terms and definitions
- Context of the organization
- Leadership
- Planning
- Support
- Operation
- Performance evaluation
- Improvement
Clauses 4 to 10 are the requirements an auditor will test. In ISO 27001 none of them can be excluded; only Annex A controls can be ruled out, with a justification. ISO 42001 doesn't spell this out as plainly, but it is built on the same logic, and certification bodies treat it the same way.
The real difference is the subject matter. ISO 27001 protects information: its confidentiality, integrity and availability. ISO 42001 governs AI systems across their lifecycle, whether you develop them, provide them to others or simply use them. The second question is broader. It asks not only "is this secure?" but "is this responsible, and who could it hurt?"
Each standard also leans on its own vocabulary document: ISO/IEC 27000 for 27001, and ISO/IEC 22989 (AI concepts and terminology) for 42001. ISO 42001 adds a set of its own definitions on top.
What Carries Straight Over
Most clauses map one-to-one. The requirement is the same; only the object changes from "information security" to "AI". Where ISO 42001 asks for something extra, it is usually a small addition rather than a new process.
Clause | ISO 27001:2022 | ISO 42001:2023 | What changes for AI |
|---|---|---|---|
4.1 | Context of the organization | Same | You must also define your role toward AI: developer, provider, user, or several at once |
4.2 | Interested parties | Same | Expect new stakeholders: people affected by AI outputs, AI regulators, model and data suppliers |
4.3 | ISMS scope | AIMS scope | One scope document can cover both systems |
5.1 | Leadership and commitment | Same | No real difference |
5.2 | Information security policy | AI policy | The AI policy must reference related internal policies where relevant |
5.3 | Roles and authorities | Same | One owner can oversee both; one auditor can audit both |
6.1.1 | Risks and opportunities | Same | Actions on AI risks and opportunities must be documented |
6.1.2–6.1.3 | Security risk assessment and treatment | AI risk assessment and treatment | Consequences are judged for individuals and society, not just the organization (see below) |
6.2 | Security objectives | AI objectives | Same mechanics, different targets |
6.3 | Planning of changes | Same | No real difference |
7.1–7.4 | Resources, competence, awareness, communication | Same | Extend training to AI-specific skills; one training plan can serve both |
7.5 | Documented information | Same | One document-control procedure covers both |
8.1 | Operational planning and control | Same | No real difference |
8.2–8.3 | Performing risk assessment and treatment | Same, for AI | Run on the same cycle as your security risks |
9.1–9.3 | Monitoring, internal audit, management review | Same | Shared audit procedure; review inputs must cover each standard separately |
10.1–10.2 | Improvement, corrective action | Same | One corrective-action procedure serves both |
If your ISMS documentation is in good shape, the practical job here is to extend existing documents, not write new ones. Add AI to the scope statement, the requirements register, the training plan and the management review agenda, and most of clauses 4 to 10 are covered.
What is genuinely new: the AI system impact assessment
The single biggest gap between the two standards is the AI system impact assessment. ISO 27001 has nothing like it. ISO 42001 requires it twice: clause 6.1.4 says you need a defined process for it, and clause 8.4 says you must actually run it at planned intervals and whenever significant changes are proposed or happen.
A security risk assessment asks what could go wrong for the organization. An impact assessment asks what an AI system could do to people outside it. It covers individuals, groups and society as a whole, and it looks at the full picture:
- how the system is developed, deployed and provided
- what it is intended to do
- how it could reasonably be misused
The results must be documented, fed into your AI risk assessment, and made available to relevant interested parties. That last point matters. This is not a private internal worksheet; it is evidence you may need to show regulators, customers or the people affected.
The quieter changes to risk management
AI risk assessment under ISO 42001 looks familiar, but four details catch ISMS teams out:
- Wider consequences. When you rate the impact of a risk, you must consider individuals and society, not only business loss.
- Two annexes to choose controls from. Treatment decisions must consider both Annex A (the controls) and Annex B (how to implement them).
- Necessity and relevance. Each control you select must be justified against the risk treatment option you chose.
- A separate Statement of Applicability. Your AI SoA is built against ISO 42001 Annex A only. It does not replace or merge into your ISO 27001 SoA.
The annexes: where the standards really diverge
The clauses are near-twins; the annexes are not. ISO 27001 has one annex. ISO 42001 has four, and three of them have no counterpart in 27001.
Annex | ISO 27001:2022 | ISO 42001:2023 |
|---|---|---|
A | 93 information security controls in 4 themes | 38 AI controls in 9 areas, focused on AI-specific risk |
B | None (the equivalent guidance sits in a separate standard, ISO/IEC 27002) | Implementation guidance for each Annex A control, built into the standard |
C | None | Example AI-related objectives and risk sources to feed your risk assessment |
D | None | How an AIMS applies across sectors, and how it integrates with ISO 27001, ISO 27701 and ISO 9001 |
Two practical consequences. First, your existing security controls do not satisfy ISO 42001 Annex A; the two control sets overlap only in places (data handling and suppliers, for example), so expect to build a largely new control library. Second, you get more help out of the box. Annex B does for AI what ISO 27002 does for security, so you don't need to buy a companion standard to understand what each control means. For deeper AI risk methodology, Annex C points to ISO/IEC 23894, the guidance standard on AI risk management.
Running both as one system
The worst way to add ISO 42001 is to build a second management system next to the first. You end up with two policies, two audit calendars and two management reviews that say the same things. Integrate instead.
Share these outright:
- Scope document, covering both the ISMS and the AIMS
- Procedure for identifying interested parties and legal, regulatory and contractual requirements, plus one requirements register
- Document and record control procedure
- Training and awareness plan
- Internal audit procedure and audit programme
- Corrective action and continual improvement procedure
Combine, but keep the inputs distinct:
- Policy: one document can hold both, but the AI commitments must be explicit and point to related policies
- Risk methodology: one method, with AI-specific impact criteria added
- Management review: one meeting and one set of minutes, with separate inputs for each standard
Build new:
- AI system impact assessment process and records
- AI risk register (or AI columns in your existing register)
- Statement of Applicability against ISO 42001 Annex A
- AI controls drawn from Annex A, implemented with Annex B guidance
The bottom line
ISO 27001 gives you the engine: governance, risk management, documentation, audit and improvement. ISO 42001 runs on that same engine but points it at a harder question, which is what your AI systems could do to people, not just to your data.
If you already hold ISO 27001, treat ISO 42001 as an extension project, not a fresh start. Reuse the management system, invest your effort in the impact assessment and the new control set, and run one integrated audit and review cycle.
Want to go deeper on AI governance each week?
Subscribe to my newsletter, AI Governance – Decoded: a 15-minute read that takes you from zero to confident on AI governance in six months.
Kumail Mehdi
ISO 42001 Lead Auditor and AI strategist. 11 years in corporate leadership, 14 years running a digital agency. I help professionals, consultants and AI startups turn expertise into governed, AI-powered systems.
