ISO/IEC 42001 is a management system standard for organisations that build or use AI. It asks you to set policy, assign roles, assess AI risks and impacts, run controls and improve over time. A first security hire can start with an inventory of AI uses, named owners and a short review cycle.
What is ISO 42001?
It is an international standard for an AI management system, in the same family of management standards as ISO 27001 for information security. It is certifiable by an accredited body. Like other management system standards it follows a plan, do, check and act cycle: define scope and context, set objectives, operate controls, audit internally and have management review the results. It also comes with a list of reference controls for AI specific topics such as impact assessment, data quality, transparency and third party use.
Read the current text through ISO or your national standards body. The ISO page could not be fetched automatically during our research, so confirm details such as clause names before you quote them.
Why would you care?
Customers and regulators increasingly ask how you govern AI. The EU AI Act sets legal obligations, and a management system is one way to organise the evidence, although a certificate does not equal legal compliance. Some buyers now include AI governance questions in security reviews.
What does a first security hire do in 90 days?
Days 1 to 30: find out what exists. Ask every team where AI is used: customer facing features, internal copilots, coding assistants, vendor tools with AI inside. Write one line per use with owner, data involved, model or vendor, and whether customers are affected. A constructed finding: a support team pastes tickets into a public chatbot, and nobody in security knew.
Days 31 to 60: set the rules. A short AI use policy covering approved tools, forbidden data types, human review for consequential decisions and how to request a new tool. Name an accountable executive and a working group of engineering, legal and product.
Days 61 to 90: add a light risk process. For each higher impact use, record the intended purpose, who could be harmed, how outputs are checked, and how you would notice drift or abuse. Review quarterly.
How does it relate to other standards?
It sits beside ISO 27001, where both use the same management cycle, and beside the NIST AI Risk Management Framework, which is voluntary guidance with different structure. You can use NIST's functions to design controls and ISO 42001 as the wrapper for audit evidence. Be clear about which is which in documents.
Where do people go wrong?
- Writing a long policy no one reads.
- Treating a certificate as the goal and skipping the inventory.
- Leaving third party AI features out of scope.
- Ignoring security issues specific to AI, such as prompt injection, data leakage through tools and over permissioned agents.
What to do
- Run the AI inventory survey and keep the result in one register.
- Write a one page AI use policy and get executive approval.
- Name owners for each AI use and an accountable executive.
- Create an impact and risk checklist for higher risk uses.
- Add AI vendors to your supplier reviews.
- Schedule a quarterly review and record decisions.
- Decide about certification only after the basics work.
Quick check
Is ISO 42001 required by law?
No. It is a voluntary standard. Laws such as the EU AI Act are separate, and certification does not by itself show legal compliance.
Where should we start?
With an inventory of where AI is used and who owns each use. Everything else in the standard builds on knowing that.
Sources
Last reviewed: 18 October 2026.
