ISO 42001 Readiness Guide: Building a Certifiable AI Management System

ISO 42001 (formally ISO/IEC 42001:2023, published in December 2023) is the international standard for an AI management system: the set of policies, processes, roles and controls an organization uses to develop, provide or use AI responsibly. It is certifiable, which means an accredited third party can audit you against it and issue a certificate. ISO 42001 readiness is the work of building that system and generating enough evidence that it operates.

The standard is written for any organization that develops, provides or uses AI, regardless of size or sector. For enterprises that already run ISO 27001, much of the structure will feel familiar. What is new is the AI-specific substance: impact assessments, controls over data used for AI, life cycle processes, and obligations to the people affected by your systems. This guide walks through the structure, the Annex A controls at a high level, and a practical readiness plan.

How ISO 42001 is structured

ISO 42001 uses the harmonized structure shared by ISO management system standards (sometimes still called Annex SL). Clauses 1-3 cover scope, normative references and terms. Clauses 4-10 contain the requirements an auditor will test.

Clause	Topic	What it asks of you
4	Context of the organization	Understand internal and external issues, interested parties and their requirements, your role(s) with respect to AI, and define the scope of the AI management system
5	Leadership	Top management commitment, an AI policy, and assigned roles, responsibilities and authorities
6	Planning	Address risks and opportunities, perform AI risk assessment and treatment, conduct AI system impact assessments, set AI objectives and plan changes
7	Support	Resources, competence, awareness, communication and documented information
8	Operation	Operational planning and control, and carrying out the risk assessment, risk treatment and impact assessment processes in practice
9	Performance evaluation	Monitoring and measurement, internal audit, management review
10	Improvement	Continual improvement, nonconformity and corrective action

Four annexes follow:

• Annex A (normative): reference control objectives and controls.
• Annex B (normative): implementation guidance for the Annex A controls.
• Annex C (informative): potential AI-related organizational objectives and risk sources.
• Annex D (informative): use of the AI management system across domains and sectors.

Two ideas that make ISO 42001 different

First, your role matters. The standard asks you to determine whether you are an AI provider, producer, customer, user, partner or another role for each relevant system. A company that builds an AI product for customers has different obligations than one that uses a vendor’s model for internal document search. Many enterprises are several of these at once, and the scope statement should say so.

Second, impact assessment is separate from risk assessment. Risk assessment looks at risks to the organization and its objectives. The AI system impact assessment looks at potential consequences for individuals, groups and society. An auditor will expect both, and will expect the impact assessment results to inform risk treatment.

Annex A controls at a high level

Annex A groups its controls under nine topic areas. You do not have to implement every control, but you must justify every inclusion and exclusion in a Statement of Applicability, just as in ISO 27001.

1. Policies related to AI: an AI policy, its alignment with other organizational policies, and its periodic review.
2. Internal organization: roles and responsibilities for AI, and a way for people to report concerns.
3. Resources for AI systems: documenting the data, tooling, system and computing resources, and human resources AI systems depend on.
4. Assessing impacts of AI systems: a process for impact assessment, documentation of results, and assessment of impacts on individuals, groups and society.
5. AI system life cycle: objectives for responsible development, and documented processes for design, development, verification and validation, deployment, operation and monitoring, and technical documentation, including event logs.
6. Data for AI systems: data management processes, acquisition, quality, provenance and preparation.
7. Information for interested parties: documentation and information for users, external reporting, communication of incidents, and information for other interested parties.
8. Use of AI systems: processes for responsible use, objectives for responsible use, and use according to intended purpose.
9. Third-party and customer relationships: allocation of responsibilities, suppliers, and customer considerations.

Read against an actual AI architecture, several of these controls become platform requirements rather than policy statements. Event logging, version control of models and prompts, data provenance, and monitoring are far easier to evidence when they are built into a shared AI platform than when each team documents them by hand.

ISO 42001 and ISO 27001: integrate, do not duplicate

If you hold ISO 27001 certification, you already run context analysis, leadership commitment, document control, competence management, internal audit, management review and corrective action. ISO 42001 uses the same clause structure, so these can be extended rather than rebuilt.

A practical integration pattern:

• One integrated management system, with a combined scope statement covering information security and AI, or an AI scope nested inside the existing one.
• Shared processes for document control, internal audit, management review and corrective action, with AI-specific inputs added to each.
• Separate but linked registers: the AI risk register and impact assessments sit alongside the information security risk register, with cross-references where an AI risk is also a security risk (prompt injection, data leakage through model outputs, model supply chain).
• A combined audit program, often with the same certification body conducting both audits, which can reduce disruption.

Be careful not to assume ISO 27001 controls cover AI risks. Access control and encryption do not address confabulation, harmful bias, or an agent taking actions outside its intended purpose. The AI-specific controls in Annex A exist because those gaps are real.

ISO 42001 readiness: a step-by-step plan

Step 1: define scope and roles

Decide which organizational units, AI systems and roles are in scope. A narrow first scope, such as one product line that provides AI to customers, can be certified faster and extended later. Scope too narrow, though, and the certificate will not answer the questions your customers are asking. Document your role (provider, user, or both) for each system.

Step 2: build the AI system inventory

You cannot scope, assess or control systems you have not found. Build the inventory from procurement records, cloud billing for model services, API gateway logs, code repositories and self-declaration. Record the owner, purpose, model and vendor, data used, and user population for each.

Step 3: write the AI policy and assign accountability

Top management must approve an AI policy that includes a commitment to meet applicable requirements and to continual improvement. Name who owns the management system, who approves AI risk treatment, and who can halt a system.

Step 4: establish risk and impact assessment methods

Define your risk criteria, likelihood and consequence scales, and acceptance thresholds. Design the impact assessment template separately, covering affected people, potential harms, severity, reversibility and mitigations. Annex C lists potential objectives and risk sources and is a good prompt list. Many organizations borrow structure from the NIST AI RMF here; see our NIST AI RMF implementation guide.

Step 5: select controls and write the Statement of Applicability

Go through Annex A, decide which controls apply, map each to an owner and to evidence, and justify exclusions. Add controls beyond Annex A where your risk assessment calls for them.

Step 6: embed controls in the AI life cycle and platform

This is where readiness succeeds or fails. Controls should live in the way systems are actually built and run: an intake step that triggers impact assessment, a release gate with required evaluation evidence, automatic logging of model calls, version pinning for models and prompts, data lineage for training and retrieval sources, and monitoring with named owners. Our enterprise AI reference architecture guide covers where these components fit.

Step 7: operate, then audit yourself

Run the system for a meaningful period before the certification audit. Complete at least one full internal audit covering all clauses and applicable controls, raise and close nonconformities, and hold a management review with the required inputs. Auditors look for evidence that the system operates, not only that documents exist.

Step 8: choose a certification body and plan the audit

Select a certification body accredited for ISO 42001 by a recognized accreditation body (ANAB in the United States, UKAS in the United Kingdom, and equivalents elsewhere). Certification follows the usual ISO pattern: a stage 1 audit reviews documentation and readiness, and a stage 2 audit tests implementation and effectiveness. Certificates are typically maintained through periodic surveillance audits and a recertification audit on a three-year cycle.

A readiness self-check

Before engaging an auditor, you should be able to answer yes to each of these:

1. Is the AIMS scope documented, including the AI systems and your role for each?
2. Is there an approved AI policy and a named owner for the management system?
3. Is the AI system inventory complete for the scope, and is it kept current?
4. Have risk assessments and impact assessments been completed for in-scope systems?
5. Is there a Statement of Applicability with justified inclusions and exclusions?
6. Do release decisions for AI systems follow a documented gate with retained evidence?
7. Are model and data versions, logs and monitoring results retained and retrievable?
8. Are supplier responsibilities for AI defined in contracts?
9. Has an internal audit been completed and have findings been addressed?
10. Has top management held a management review with the required inputs?

Common readiness mistakes

• Documentation-first programs. Writing every procedure before any system runs through them. Auditors find the gap quickly.
• Scope that excludes the risky systems. Certifying the easy parts undermines the point of certifying at all.
• Treating vendor models as out of scope. Using a third-party model does not remove your obligations as a user; Annex A has controls specifically for third-party relationships.
• Impact assessment as a one-time form. Impacts change when models, data or intended use change. Trigger reassessment on change.
• No technical evidence. Claiming monitoring and logging without systems that produce the records.

How we help

We prepare organizations for ISO 42001 by designing the management system and building the controls into the AI platform itself, so the evidence is produced by normal operation. Our AI governance and risk practice leads readiness work and integrates it with your existing ISO 27001 program. Contact us to scope a readiness assessment.