Skip to content
Diagram of ISO 42001 clauses 4 to 10 arranged around a central AI policy

ISO 420018 min read

ISO 42001 requirements explained for small businesses

What ISO 42001:2023 asks of a business that builds or uses AI: clauses 4 to 10, Annex A controls, required documents and a checklist to work through.

By Michael Schwarz, Lead auditor ISO 9001 and 14001, SICE

ISO 42001:2023 requires you to set up, run and keep improving a management system for artificial intelligence. That means a written AI policy, a defined scope, a risk and impact assessment for your AI systems, controls for how they are built, bought and used, and proof that you check all of this regularly.

The standard does not tell you which AI tools to use or how accurate a model must be. It describes how an organisation governs AI. If you already know ISO 9001 or ISO 27001, the structure will look familiar.

What ISO 42001 covers

ISO 42001 is the first international management system standard written for AI. It applies to organisations that develop AI, provide AI-based products or services, or use AI systems in their daily work. A ten-person agency that uses a language model for client copy is in scope in the same way as a software company that trains its own models. The size of the effort differs, the logic does not.

The official description of the standard is on the ISO page for ISO/IEC 42001. The full text is sold by ISO and national standards bodies. This article explains what the text asks for, not the wording itself.

The standard follows the high level structure shared by other ISO management system standards. Clauses 4 to 10 contain the requirements you must meet. Annex A lists AI-specific controls, and Annex B gives guidance on applying them.

The requirements clause by clause

Clause 4: Context of the organization

You describe where your organisation stands with AI. Which AI systems do you build, buy or use? Who is affected by them: customers, employees, applicants, the public? Which laws, contracts and customer expectations apply?

From this you define the scope of the management system. A small business can scope the whole company or a single service, for example the chatbot it runs for clients. The scope must be written down and must match what you actually do.

Clause 5: Leadership

Top management owns the system. In a small company that is the owner or managing director. They approve an AI policy, assign roles and responsibilities, and make sure the system gets the time and budget it needs.

The AI policy is a short document. It states what you use AI for, which principles you follow and who is responsible. Employees must be able to find it and know what it means for their work.

Clause 6: Planning

This is the core of the standard. You identify risks and opportunities, and you assess the risks that come from your AI systems: wrong outputs, bias, leaked data, dependence on a single provider, misuse by staff. You decide how to treat each risk and which Annex A controls you apply.

ISO 42001 also asks for an assessment of the impact of an AI system on people and society, beyond the risk to your own business. A tool that screens CVs affects applicants. A tool that drafts a contract summary affects the client who relies on it. The assessment records who could be harmed and how.

You keep a statement of applicability. It lists the Annex A controls, says which ones you apply and gives a reason for each one you leave out. Finally you set measurable AI objectives and plan how to reach them.

Clause 7: Support

You provide the people, skills and information the system needs. Staff who work with AI must be competent for it, which means training or documented experience. They must also be aware of the policy and of what happens if they ignore it.

This clause also covers communication, internal and external, and documented information. Documents must be identifiable, approved, current and protected from accidental change.

Clause 8: Operation

Here the plans become daily practice. You carry out the risk assessments and impact assessments at planned intervals and whenever something changes, for example a new model version or a new use case. You apply the risk treatment, and you control the AI system across its life cycle: design or selection, testing, deployment, monitoring and retirement.

If a supplier provides the AI, you still need to manage it. That means knowing what the supplier tells you about its system, what data goes into it, and what you do when the supplier changes terms.

Clause 9: Performance evaluation

You measure whether the system works. This includes monitoring of your AI objectives and risks, an internal audit at planned intervals, and a management review in which top management looks at results, incidents and open actions.

The internal audit must be carried out by someone who did not write the thing being audited. In a small company this is often the hardest part to arrange, and it is acceptable to use an external person for it.

Clause 10: Improvement

When something goes wrong, you record it, correct it and look for the cause so it does not return. A nonconformity could be an AI output that reached a client without review although your rule requires review. You also improve the system on an ongoing basis, not only after incidents.

What Annex A adds

Annex A is a catalogue of controls grouped by topic. The topics include policies for AI, internal organisation and roles, resources for AI systems, assessment of impacts, the AI system life cycle, data for AI systems, information for interested parties, responsible use, and relationships with third parties and customers.

You are not required to apply every control. You are required to consider all of them in your risk treatment and to justify exclusions. For a business that only uses third-party AI tools, controls about design and development of models will often be excluded with a short reason. Controls about data handling, human oversight and supplier management will almost always stay.

Documents you need to be able to show

The standard asks for documented information in several places. In practice a small business needs these:

  • the scope of the AI management system
  • the AI policy
  • an inventory of AI systems in use or in development
  • the AI risk assessment and risk treatment results
  • the AI system impact assessments
  • the statement of applicability
  • AI objectives and the plan to reach them
  • evidence of competence and training
  • internal audit report and management review minutes
  • records of incidents, nonconformities and corrective actions

None of these needs to be long. Several short, accurate documents are better than a thick manual nobody reads.

Checklist: are you ready for ISO 42001?

If you can tick most of these, the remaining gaps are usually documentation, not substance.

Common mistakes small businesses make

The first is scoping too wide. A company that lists every conceivable AI use in a single scope ends up with controls it cannot evidence. Start with the systems that matter to clients.

The second is ignoring shadow use. Employees paste client text into public chatbots long before management writes a policy. The inventory must reflect reality, not the approved tool list.

The third is treating the impact assessment as a copy of the risk assessment. Risk looks at what can hurt your business. Impact looks at who else can be harmed. The two overlap, but they are separate records.

The fourth is a policy that sits on a shared drive. An auditor will ask an employee what the policy says. If nobody knows, the requirement of clause 7 on awareness is not met.

Certificate options for ISO 42001

The classic route is an on-site audit by a certification body, with a stage 1 and stage 2 audit and surveillance visits. It suits organisations whose customers or regulators name that route specifically. A shorter path exists for businesses that need to show a purchasing department or a tender team that their AI governance has been assessed. Our guide on ISO certification without an audit explains the difference, and the page on ISO certification for small and medium businesses covers how companies of your size usually approach it.

How Zertify fits in

Zertify issues an ISO 42001 certificate based on an online self-assessment. There is no on-site audit. You answer questions that follow the requirements described above, and you upload no documents you do not already have. A member of the SICE Certification Board reviews every assessment. After a passed assessment and payment, Zertify issues the certificate within 4 hours, under licence from SICE.

The certificate carries a QR code that leads to a public verification page, so a customer can confirm it without calling you. It is valid for three years and is confirmed each year with a short online re-audit, which is a yearly subscription.

The details of what is assessed are on the ISO 42001 page. Costs are on the pricing page. When your inventory, policy and risk records are in place, you can begin on the assessment start page.

A certificate shows that you run an AI management system. It does not replace legal advice on the laws that apply to your AI use.

Frequently asked questions

What are the main requirements of ISO 42001?

You define the scope and context of your AI use, get top management to approve an AI policy, assess risks and impacts of your AI systems, apply and justify controls from Annex A, train staff, operate the controls across the AI life cycle, audit internally, hold a management review and correct nonconformities.

Does ISO 42001 apply if we only use AI tools and do not build them?

Yes. The standard covers organisations that use AI systems as well as those that develop or provide them. If you only use third-party tools, many design and development controls can be excluded with a reason, while data handling, oversight and supplier controls remain relevant.

Do we have to apply every Annex A control?

No. You must consider every control in your risk treatment and record in the statement of applicability which ones you apply. Each exclusion needs a stated reason.

Which documents does ISO 42001 require?

Among others: the scope, the AI policy, risk assessment and treatment results, AI system impact assessments, the statement of applicability, objectives, evidence of competence, internal audit results and management review records.

How long is a Zertify ISO 42001 certificate valid?

Three years. It is confirmed every year with a short online re-audit, which is a yearly subscription.

Keep reading

More articles

ISO 42001by Michael Schwarz

ISO 42001 Checklist for an AI Management System

A practical ISO 42001 checklist: scope, AI inventory, policy, risk and impact assessment, Annex A controls, monitoring and improvement, in the order you work.

Checklist with ticked boxes next to a simple diagram of an AI system and its data flows

7 min read

An ISO certificate for your business

Complete the self-assessment online and see your result straight away. You pay only after you pass.

Certificate usually within 4 hours of payment. Issued under licence from SICE, verifiable by QR code.