A client questionnaire has started asking about AI, or someone on your board has read about ISO 42001, and the question arrives as a choice: do you need the new AI standard, or is ISO 27001 enough? For most businesses that use AI tools rather than build them, the answer is ISO 27001 first, extended to cover AI, with ISO 42001 later if a client asks for it by name.
Cardonet is certified to ISO 27001 and ISO 9001, and we do not hold ISO 42001. Because we have a reason to like the first answer, what follows sets out the test for you to apply yourself.
What each one covers
ISO 27001 is the international standard for managing information security. It asks you to know what information you hold, assess the risks to it and run controls that keep it confidential, accurate and available, with an independent auditor checking every year.
Although ISO 27001 says nothing about AI by name, it does not need to for most of what matters. An AI tool is a supplier handling your information and an account with access to your systems, and the standard already has controls for both.
ISO 42001, published at the end of 2023, is the first international standard for managing AI itself. It covers what ISO 27001 does not: the effect an AI system has on the people its output touches, whether a person oversees it, how its data is chosen and how it is looked after from first use to retirement.
Because its scope depends on your role, a business that builds AI into a product carries more of it than one that uses tools from a supplier.
Both are certified by an accredited certification body, and both are built on the same clause structure ISO uses for all its management system standards. A business that already runs one has the risk assessment, internal audit and management review that the other needs, which makes adding the second a smaller project than starting from nothing.
The question each one answers
ISO 27001 answers "is our information safe when it goes into this tool?". ISO 42001 answers "is the AI itself under control: what it does, who it affects and who checks it?". If a business only uses AI tools from suppliers, it is mostly asking the first question. A business whose own product makes decisions about people is asking both.
| ISO 27001 | ISO 42001 | |
|---|---|---|
| What it manages | Information security | AI itself |
| What it covers | What information you hold, the risks to it and the controls that keep it confidential, accurate and available | The effect on the people an AI system's output touches, human oversight, how its data is chosen and how it is looked after from first use to retirement |
| The question it answers | Is our information safe when it goes into this tool? | Is the AI itself under control? |
| Who certifies it | An accredited certification body | An accredited certification body |
| When it fits | Usually enough when you use AI tools from suppliers | Worth it when you build AI into a product, use AI to make decisions about people or a client asks for it by name |
When ISO 27001 is enough for now
It is usually enough when all three of these hold:
- You use AI tools from suppliers, such as Microsoft 365 Copilot, ChatGPT Business or Claude.
- Your exposure is what goes into those tools and what they can reach, which is an information security question.
- Clients ask how you control AI use without naming a certificate.
If you already hold ISO 27001, extending it to cover AI is ordinary work inside the system you run, and what ISO 27001 expects once AI is in use goes through it control by control.
AI tools go onto the asset and supplier registers, the risk assessment covers them, each supplier is assessed like any other, and acceptable use and awareness training cover what may go in. Access is reviewed before a tool is connected, for the reasons set out in what Copilot can already see. The written rules are the ones in what an AI policy should contain.
When ISO 42001 is worth it
- You build AI into a product or service your clients use, or you use AI to make decisions about people, such as screening candidates or assessing eligibility.
- A client, a tender or a regulator asks for it by name.
- You provide an AI system into the European Union that may fall into the AI Act's high-risk categories. ISO 42001 is not the European harmonised standard for the Act, and as of October 2026 the European standards that would give a presumption of conformity were not yet final, so take legal advice on what applies to you.
- You already run ISO 27001 and want one management system that covers AI as well, which is where the shared structure pays off.
What your suppliers' certificates cover
Several of the vendors you are likely to use hold ISO 42001. Anthropic was certified in January 2025, and Microsoft 365 Copilot was recertified in 2026 after first certifying in March 2025. For the supplier section of a client questionnaire, that is useful evidence, and it tells you the tool is built and run under an audited system.
It does not cover how your business uses the tool. Your rules on what may go in, who is connected, what each person can reach and who reviews the output before it leaves are yours, and no supplier's certificate answers for them.
What drives the cost of either
There is no single figure, because the drivers differ so much between businesses. They are the scope (how many systems and which roles are covered), whether you already run a management system and how much internal time goes into risk and impact assessments. For the impact assessment part, ISO/IEC 42005, published in May 2025, gives guidance. The certification body's fees come on top: an initial two-stage audit, a check every year and a full recertification every three years.
Where we fit
We build the controls and the evidence that either standard asks a business to show for its AI use: the tool register, the written policy, sign-in under your company account, the permissions review and the record of who approved what.
That work sits in Govern, and Guardrails lists what we do and do not claim. Although we run our own work under ISO 27001 and ISO 9001, we do not certify anyone, which only an accredited certification body can do, and we do not give legal advice.
If you would like a first view of where you stand, ten questions take about two minutes and send nothing anywhere.