AI Law, Risk & Governance

ISO/IEC 42001: The AI Management System, Effort, Benefits and the Route to Certification

Marin Niehues18 May 202612 min read

A screen with metrics and trend lines

ISO/IEC 42001 is the international standard for AI management systems and sets out how an organization governs the development and use of AI and can have this certified. This article explains the structure and benefits of the standard, calculates the effort and describes the route to a certificate. This article provides an overview and does not constitute legal advice.

What is ISO/IEC 42001?

ISO/IEC 42001 is the first international standard that specifies requirements for a management system for artificial intelligence. ISO and IEC published it jointly in 2023. It applies to any organization that develops, provides or uses AI systems, regardless of sector and size.

A management system is the documented framework an organization uses to govern a topic on a lasting basis, with a leadership policy, named roles, a risk process and a cycle of review and improvement. Companies know the principle from ISO/IEC 27001. ISO/IEC 42001 applies it to AI and adds questions specific to AI, such as the impact of a system on people and the provenance of data.

What gets certified is the organization's management system. An individual model receives no seal of approval, and the certificate says nothing about the accuracy of a particular system.

The benefit lies in speed and the ability to provide evidence. Every review has a fixed place in the process, new applications pass through a known procedure, and AI risks land in the same register as all other corporate risks. How to build that register is covered in AI risk management: a framework for decision-makers, and the article AI management: what executive leadership must steer places it within the wider steering of AI initiatives. Related standards complement ISO/IEC 42001: ISO/IEC 42005 describes the impact assessment, ISO/IEC 42006 the requirements for certification bodies.

Structure: ten clauses and a catalog of 38 controls

The standard follows the harmonized structure shared by all ISO management system standards. Clauses 4 to 10 contain the mandatory requirements, from context and leadership through planning and operation to internal audit, management review and improvement. Anyone already running a management system under ISO/IEC 27001 will recognize the structure and can reuse documents and processes.

The practical core lies in clause 6. Here the standard requires two assessments that a company should keep apart. The risk assessment asks what can go wrong for the organization. The impact assessment asks how an AI system affects people and society. An application that pre-screens job applications shows the difference: the risk for the company is litigation, the consequence for those affected is an unfair rejection.

Annex A contains 38 controls in nine areas:

  • policies related to AI
  • internal organization and roles
  • resources for AI systems
  • assessing the impacts of AI systems
  • the AI system life cycle from requirement to retirement
  • data for AI systems, with provenance and quality
  • information for interested parties
  • use of AI systems
  • third-party and customer relationships

The company selects the controls that treat its risks and justifies its choices in the statement of applicability. A company that only buys AI excludes parts of the life cycle controls, a provider of AI solutions implements almost all of them.

How does ISO/IEC 42001 relate to the AI Act?

ISO/IEC 42001 is a voluntary standard, the AI Act a binding EU law, and a certificate under the standard does not automatically fulfill the obligations of the law. The AI Act ties obligations to the risk class of a system and to the company's role as provider or deployer. The standard describes a management system for all AI systems of an organization, regardless of their risk class.

Three rules follow for practice:

  1. The management system organizes the obligations. Many requirements of the AI Act need exactly what the standard builds. These include an inventory of systems with a risk assessment, named owners and training records for the required AI literacy.
  2. Classification under the law remains a separate task. Whether a system counts as high-risk is not answered by the standard. This check belongs in the approval procedure as a fixed step.
  3. High-risk systems carry additional requirements. Providers of such systems need, among other things, a quality management system and technical documentation as specified by the regulation. The European standardization organizations are developing dedicated harmonized standards for this. ISO/IEC 42001 is a sound basis, yet it does not fully cover the requirements for the individual product.

The article on the EU AI Act gives an overview of risk classes and roles. How the reviews fit into a lean framework with a committee and approvals is described in Building AI governance.

Does a company need ISO/IEC 42001 certification?

There is no legal obligation to certify, and a certificate pays off for companies that need evidence for third parties or already run a certified management system. For everyone else, using the standard as the structure for their own governance is the more economical route. The decision follows five criteria.

The rule is this: if two criteria in the column “Pursue certification” apply, certification pays off. If only “Number and risk” applies, the company builds the system according to the standard and defers certification until a customer asks for it.

How does certification work and how long does it take?

Certification follows a two-stage audit by an accredited certification body, and with an existing management system the route to a certificate typically takes four to six months, without one nine to twelve months. Preparation takes the largest share, the audits themselves last a few days.

The route has six steps:

  1. Scope and inventory: The company defines which units and AI systems the management system covers and records each system with its purpose and owner.
  2. Gap analysis: Comparing existing rules with clauses 4 to 10 and the control catalog shows what is missing.
  3. Build: AI policy, roles and the procedures for risk and impact assessment are created or embedded in existing processes.
  4. Operation with evidence: The system runs for some months so that the audit can examine processes in actual use.
  5. Internal audit and management review: Both must have taken place at least once before certification.
  6. Certification audit: In stage 1 the certification body reviews documentation and readiness, in stage 2 implementation in operation.

The certificate is valid for three years. Annual surveillance audits take place during this period, and recertification follows before expiry. When choosing a certification body, accreditation matters, in Germany by the national accreditation body DAkkS, in the UK by UKAS. Only an accredited certificate is reliably recognized in tenders.

The effort can be calculated in person-days upfront

The effort for an AI management system consists of a fixed base effort for the system and a variable share per AI system. Estimating both parts separately gives a reliable budget. The calculation has four items:

  1. Base effort: Policy, roles and procedures, plus internal audit and management review. An existing ISO/IEC 27001 system reduces this item considerably.
  2. Effort per AI system: Risk and impact assessment with documentation. This item grows with the number and risk of the systems.
  3. External support: In the German consulting market it is typically billed at around 1,000 to 2,500 euros per consultant day, depending on seniority.
  4. Certification body: The body sets the audit duration based on headcount and complexity. Quotes from two or three accredited bodies provide clarity.

A worked example with assumed values shows the logic. A company certified to ISO/IEC 27001 plans 20 internal person-days of base effort and runs twelve AI systems at two person-days each. With 44 internal person-days and around 15 external consultant days, the build is costed, plus the audit fees. The number of AI systems is the strongest lever on the budget.

Three certificates matter when selecting a vendor

An AI vendor should hold ISO/IEC 27001 certification, additionally ISO/IEC 42001 for the governance of its AI, and for cloud services in regulated environments an attestation against the BSI C5 criteria catalog.

ISO/IEC 27001 is the minimum for any vendor processing company data. C5 is the cloud criteria catalog of the German Federal Office for Information Security. Check the scope of every certificate: if it covers only a data center, it says little about the AI service on offer. Further questions for the review are covered in the article on information security for generative AI.

Niehues Consulting as partner for the AI management system

Niehues Consulting supports executive teams in building an AI management system that works in operation and follows the structure of the standard. The focus is on AI management: inventory and classification of AI systems, decision paths and approval procedures, and the link to portfolio steering. Marin Niehues is a management consultant with a delivery perspective. He builds organizations that deliver and brings AI into operation. His projects range from VC-funded scale-ups to DAX 40 corporations, he has worked in 20+ countries on six continents, and his home base is the DACH region.

The approach follows the Niehues Delivery Model with the phases Understand, Build and Embed. Three of its review questions matter most for an AI management system. Who decides, who delivers, who uses? The answer maps onto the roles in the standard. Which roles, decision paths and interfaces does operation need? This produces the approval procedure. Who owns operation from day one? This question decides whether the system keeps running after the audit. The entry point is a free 30-minute initial consultation via the contact form.

Frequently asked questions

What is ISO/IEC 42001 in simple terms?

ISO/IEC 42001 is an international standard that describes how a company governs its use of AI on a lasting basis with rules, roles and risk assessment. It works like ISO/IEC 27001 for information security, applied to AI. Companies can have their implementation certified by an independent body.

Is ISO/IEC 42001 certification mandatory?

No, certification is voluntary, and no law requires it. It becomes mandatory in practice when customers or tenders make it a condition for a contract.

What is the difference between ISO/IEC 42001 and ISO/IEC 27001?

ISO/IEC 27001 governs information security, ISO/IEC 42001 the governance of AI including impact assessment and the system life cycle. Both share the same structure, so an existing ISO/IEC 27001 system can be extended with the AI requirements. Many companies run both in an integrated management system with a joint audit.

Does ISO/IEC 42001 certification ensure AI Act compliance?

No, a certificate does not automatically fulfill the legal obligations. It does provide the structure for organizing many of them, such as the inventory and documentation. Classifying each system under the AI Act remains a separate step.

How much does ISO/IEC 42001 certification cost?

The cost comprises internal effort, external consulting and the fees of the certification body. Internal effort grows with the number of AI systems. In the German market, external consulting is typically billed at around 1,000 to 2,500 euros per consultant day.

Is ISO/IEC 42001 worth it for mid-sized companies?

For a mid-sized company that offers AI solutions or is asked for evidence by major customers, certification pays off. A company that only uses purchased AI internally builds its governance along the structure of the standard and postpones the audit. A lean system with an inventory and an approval procedure can be certified later.

What is a statement of applicability?

The statement of applicability records which of the 38 Annex A controls a company implements and which it excludes with justification. It is a mandatory part of the management system and a central audit document.

Let’s talk about your initiative.

In 30 minutes we clarify where your organization stands and which step will pay off most now.

Request a consultation

30 minutes · free · no obligation