Implementing ISO 42001: a roadmap from AI inventory to certification

ISO/IEC 42001:2023 is the international standard for an AI management system, the AIMS. It follows the same high-level structure as ISO 27001: context, leadership, planning, support, operation, performance evaluation and improvement. Anyone who knows an ISMS will recognize the skeleton. The content differs: the risks concern the organization itself and also the consequences of AI for people and groups outside it. This roadmap follows the order in which an implementation tends to work best in practice, with the relevant clause for each step.

Step 1: map the AI you have

Everything in the AIMS depends on the list of AI systems the organization develops, provides or uses. That list is almost always longer than expected. Besides in-house models and a centrally purchased chatbot, it covers AI features in CRM, HR software, customer service tools and office suites, often switched on by a vendor without anyone deciding to use them. So do not only ask IT. Ask procurement, HR, marketing and the departments that sign up for tools themselves. How to find the hidden applications is covered in taking inventory of shadow AI. For each system, record at least the purpose, the owner, the vendor, the data used and the people affected.

Step 2: determine context, role and scope

Clause 4 requires you to determine the context of the organization, the interested parties and their requirements, and your role with respect to AI systems. Do you develop AI, provide it to customers, use it yourself, or a combination? That role determines which controls are relevant and lines up with the distinction between provider and deployer in the EU AI Act. Then you fix the scope. A scope that is too narrow, for example one product while customers ask about the entire service, results in a certificate that says little. A scope that is too broad makes the first project unnecessarily heavy.

Step 3: set leadership, policy and roles

Clause 5 requires top management commitment, an AI policy and assigned responsibilities. A common mistake is to make the security department owner of all AI risks. Every AI initiative then runs through one department that cannot carry the load. It works better when the business owns the AI system and its outcomes, while security, privacy and compliance advise and review. Keep the policy workable. A ban on everything not explicitly approved pushes employees to personal accounts, and out of sight of the AIMS. Also set measurable AI objectives (clause 6.2). An auditor will ask how you track them.

Step 4: assess and treat the AI risks

Clause 6.1.2 requires a defined method for AI risk assessment. The risks are broader than in information security: bias, lack of transparency, poor data quality, model drift, misuse and dependence on a single model provider all belong there. Make risks concrete with scenarios. What happens if this model makes a wrong decision, who notices, and what does it mean for customers, affected people and the organization? A risk register where everything is rated "medium" helps nobody prioritize.

For risk treatment (clause 6.1.3) you select controls and compare them with the 38 controls in Annex A. You record the result in a Statement of Applicability, with the reason for including or excluding each control. Annex A is a reference list, not a mandatory checklist. An exclusion is allowed if you justify it. How to build that document is explained in the Statement of Applicability for ISO 42001.

Step 5: carry out AI system impact assessments

Clause 6.1.4 has no counterpart in ISO 27001. You define a process to assess the potential consequences of an AI system for individuals, groups and society, and you document the results. Those results feed into the risk assessment. ISO/IEC 42005:2025 gives guidance on how to set up such an assessment. It is not a certifiable standard, but it is a useful template. For deployers of certain high-risk systems, the AI Act also requires a fundamental rights impact assessment (Article 27). Align the two so you do not carry out the same analysis twice.

Step 6: implement controls and build evidence

Now the AIMS starts running. Clauses 7 and 8 require you to arrange resources, competence, awareness, communication and documented information, and to carry out the planned processes. In practice this includes a life cycle process for AI systems (Annex A.6), agreements on the data used to train and feed systems (A.7), information for users and affected people (A.8), rules for responsible use (A.9) and requirements for suppliers and customers (A.10). An auditor wants to see that a procedure is followed; the document alone is not enough. So collect evidence from day one: approvals, test results, recorded assessments of new systems and records of human oversight.

Step 7: internal audit and management review

Clauses 9.2 and 9.3 require an internal audit and a management review. Certification bodies expect both to have been carried out at least once before the stage 2 audit takes place. The internal audit covers the full scope and all requirements, and is performed by someone independent of the work being reviewed. The management review covers fixed topics, such as audit results, progress on objectives and changes in context and risks. If you set up the internal audit as a dress rehearsal for the external audit, you find the weak spots yourself. An approach is described in the internal audit of your AIMS.

Step 8: the certification audit

The external audit has two stages. In stage 1 the certification body reviews the documentation and whether the organization is ready for the next stage. In stage 2 it tests whether the AIMS works in practice. ISO/IEC 42006 was published in 2025, with additional requirements for bodies that certify ISO 42001, including the competence of auditors and the calculation of audit time. Choose a body that is accredited for ISO 42001 or in the process of accreditation, and ask up front how long the AIMS must demonstrably have been running before stage 2 can take place. A certificate is valid for three years, with annual surveillance audits.

After certification

An AIMS goes out of date quickly if nobody maintains it. Models are retrained, vendors add AI features and new applications come in through a trial subscription. So set a fixed rhythm: update the AI inventory periodically, assess new systems before they go live and follow up on nonconformities with corrective actions (clause 10). At the surveillance audit, the auditor will ask about exactly those changes.

ISO 27001 and the EU AI Act

If you already have ISO 27001, an integrated management system is the obvious choice. Document control, internal audit, management review and corrective action can serve both standards. The AI-specific parts, such as the impact assessments and the controls in Annex A of ISO 42001, come on top.

For the EU AI Act, ISO 42001 is a good basis for governance, but not proof of conformity. The standard is not a harmonised standard under the AI Act and therefore gives no presumption of conformity. For the quality management system requirements for providers of high-risk systems (Article 17), CEN-CENELEC is working on a separate European standard, EN 18286. Under the Digital Omnibus, Regulation (EU) 2026/1744, the obligations for high-risk systems in Annex III apply from 2 December 2027 and those for AI in products under Annex I from 2 August 2028. If you start with ISO 42001 now, the management system is in place before those dates arrive.

Secure Audit supports organizations with implementing ISO 42001, from gap analysis and AI inventory to preparing for the certification audit. More about our approach is on the Compliance page.