Request a quote
Emergency SectorsCareers About us Blog Get in touch
Two colleagues at a table with a laptop between them in a warmly lit office

ISO 42001: a management system for AI, and how it sits alongside ISO 27001

ISO 42001 is the first international standard for an artificial intelligence management system. It is not about which AI you may use, but about how an organisation identifies, controls and demonstrably keeps track of AI risks. This article covers the questions that follow: do you need it, what does it ask beyond an AI policy, and how much of your ISO 27001 system can you reuse?

7 min read Written by the ITproposal team
Short answer

ISO/IEC 42001 requires a management system for AI: identifying, controlling, recording and reviewing the risks and impacts of AI systems.

For an organisation that only uses AI, a sound AI policy is often enough; the standard pays off mainly for those who build, supply or rely on AI in a critical process.

If you hold ISO 27001, you reuse the structure, audit and document control. A certificate is not proof of AI Act compliance, and we certify no one.

What ISO 42001 is, and what it is not

ISO/IEC 42001 was published at the end of 2023 and sets out requirements for an AI management system, or AIMS. Like ISO 27001 for information security, it gives no list of banned uses or approved tools. It requires you to know which AI you develop, supply or use, what risks and impacts come with it, what you do about them, and to keep that demonstrably up to date.

What it is not: a technical standard for model quality, a product mark, or a substitute for legislation. An organisation is certified, not a chatbot.

The standard builds on two earlier documents: ISO/IEC 22989 for concepts and roles, and ISO/IEC 23894 as guidance on AI risk management. If you want to understand the vocabulary of 42001, the first is the most useful.

Build, supply or use: who the standard suits

The key question up front is your role. The standard applies to any organisation that provides or uses AI, but the effort and the return differ widely.

  • You develop AI or build it into your product. This is where the standard fits best: customers want to know how you manage data, models and changes, and a certificate answers that in one go.
  • You supply a service with AI inside. Customers and procurement teams then ask the same questions. If they come back in every tender, a system pays off.
  • You use AI in a critical process. Assessing people, credit, access or safety. There the impacts weigh heavily enough to manage them systematically.
  • You use AI as an office tool. An assistant in email, meeting summaries. A clear AI policy and AI literacy is then usually enough, and a full management system is heavier than the risks.

If in doubt, start with an inventory of what is running. You need it in any case, and it shows which row you are in.

What ISO 42001 asks beyond an AI policy

An AI policy says what is allowed. The standard asks for the system that makes that policy real. The parts that make the difference:

  • An AI risk assessment. For each AI system: what can go wrong, for the organisation and for the people affected, and how likely it is.
  • An AI system impact assessment. This is new compared with ISO 27001: the consequences for individuals and groups outside your organisation, such as an unfair outcome or a decision nobody can explain.
  • A statement of applicability. Which controls from Annex A you apply, which you do not, and why.
  • The life cycle. How an AI system is designed, tested, deployed, changed and retired, and who decides at each step.
  • Data. Provenance, quality and suitability of the data a system works with or was trained on.
  • Information for interested parties. What users and affected people are told about what the system does and where it stops.
  • Suppliers. AI you buy in is covered too: what the supplier commits to, and what you still control yourself.

Alongside ISO 27001: what you reuse and what is new

ISO 42001 follows the same clause structure as ISO 27001 and ISO 9001. With a certified ISO 27001 system you do not start from zero; much of what you have ticked off in our ISO 27001 checklist carries over.

What you reuse:

  • The structure: context, leadership, planning, support, operation, evaluation, improvement.
  • Internal audit, management review, document control and corrective action.
  • The security controls around access, logging, suppliers and incidents; an AI system is an information system too.
  • Your risk method, provided it is broad enough to weigh more than confidentiality, integrity and availability.

What is new: the impact assessment on people, the model life cycle, data quality and provenance, explainability and information for those affected. ISO 27001 does not ask those questions, and they often sit with people other than the security lead: product management, legal, the process owner.

Combine the two in one integrated management system with one audit cycle. Two separate systems side by side mean holding the same management review twice.

ISO 42001 and the EU AI Act: what a certificate does and does not prove

The EU AI Act is legislation; ISO 42001 is a voluntary standard. They touch, but they are not the same.

  • What a certificate shows: that your organisation manages AI risks systematically, with a working management system assessed by an independent body.
  • What it does not show: that a particular AI system complies with the AI Act. At the time of writing, ISO 42001 is not a harmonised standard under that regulation, so a certificate gives no presumption of conformity.
  • Where it does help: the AI Act requires providers of high-risk AI to have, among other things, risk management, data governance, documentation and human oversight. A working AIMS supplies much of the evidence and the way of working you need for that.

If an AI system may fall under the AI Act, have a lawyer assess its risk category and the obligations that come with it. A management system does not answer that question for you. In the UK, which is outside the AI Act, the standard is mostly asked for by customers who sell into the EU.

From AI inventory to audit: the sequence

  • Inventory. Which AI systems you develop, supply or use, including AI built into existing software and features staff have switched on themselves.
  • Role and scope. For each system your role (developer, provider, user) and which systems fall inside the management system.
  • Policy and responsibilities. An AI policy the board stands behind, and an owner for each system.
  • Risk and impact assessment. Per system, with the people who use it and the people who feel its effects.
  • Controls and statement of applicability. What you do, what you deliberately do not, and why.
  • Operate and evidence. A few months of running it, with records, assessments and changes you can show.
  • Internal audit, management review, certification audit. The last one in two stages with a certification body. Check that the body is accredited for this particular standard, in the UK by UKAS; with a new standard that is not a given.

What drives the effort: how many AI systems are in scope, whether you build or buy them, what is already in an ISO 27001 system, and how much of your AI use is still unknown.

Where it goes wrong

  • A scope covering only your own product. The AI in your email, CRM and meeting software also falls under your policy, and an auditor will ask about it.
  • The impact assessment as a form. If nobody outside IT helps write it, exactly the consequences that matter are missing.
  • Taking suppliers at their word. “We do not train on your data” is a sentence; ask where it is in the contract and how you check it.
  • A snapshot. AI features arrive unasked with software updates. Without a fixed review the inventory is out of date within a quarter.
  • Confusing the certificate with compliance. If you must comply with the AI Act, you have to demonstrate that separately.

Where our work stops

What we do. We deliver the technical side: mapping which AI runs in your environment and what access it has, setting up AI assistants such as Microsoft 365 Copilot within existing permissions, labelling and protecting sensitive data, putting logging and access management in order, and supplying the evidence from our management that an auditor asks for on the security controls. We work under ISO 9001, ISO/IEC 27001, ISO 14001 and Kiwa NEN 4400-1 ourselves.

What we do not do. We are not a certification body and not an auditor. We do not issue an ISO 42001 certificate, we do not tell you whether you comply with the standard, and we do not carry out an independent audit. Nor are we lawyers: whether an AI system falls under the AI Act and which obligations apply is for a legal assessment. The impact assessment on people you write with your own process owners; we supply the technical facts for it.

Frequently asked

Questions we get about this

The ones that come up most when ISO 42001 is on the table.

What is ISO 42001?

ISO/IEC 42001 is the international standard setting requirements for an artificial intelligence management system (AIMS). It asks an organisation to establish which AI it develops, provides or uses, what risks and impacts come with it, and how it controls and maintains them. An organisation can be certified against it.

Is ISO 42001 mandatory?

No, ISO 42001 is a voluntary standard. Customers and tenders increasingly ask for it from suppliers that develop AI or use it in their service. Legal obligations around AI come from legislation such as the EU AI Act, not from the standard.

What is the difference between ISO 42001 and ISO 27001?

ISO 27001 is about information security; ISO 42001 is about developing and using AI responsibly, including the impact on people outside your organisation, data quality and the model life cycle. They share the same clause structure, so you can combine them in one management system.

Does an ISO 42001 certificate prove compliance with the EU AI Act?

No. A certificate shows that you manage AI risks systematically, but it does not say that a particular AI system complies with the AI Act. At the time of writing, ISO 42001 is not a harmonised standard under that regulation. A working management system does help with the evidence the AI Act asks for.

Can you get us ISO 42001 certified?

No. Only an accredited certification body issues a certificate, and we are not a certification body or an auditor. What we do is put the technical side in order: visibility of which AI is running, access rights, labels and logging, and the evidence for it from our management.

Related services

Further reading

The policy, the security beneath it and the AI itself.

Shall we start by mapping which AI is running?

Tell us which AI tools and features you know of and which processes they sit in. We will show what is technically active in your environment and what access it has; certification we leave to your certification body.

Practical IT knowledge in your inbox

New guides on management, security and the workplace, written by the people doing the work. No sales talk, and you can unsubscribe in one click.

We use your address for the newsletter only. See our privacy policy.

Request a quote Call