Getting certified to ISO 27001 is a significant piece of work, and once the certificate is in hand it’s natural to expect it to carry most security conversations with customers. For a long time, it did.
Lately, though, security reviews have started to include a section that wasn’t there a year or two ago. It asks whether your product uses AI. Which models. Whether the customer’s data is used for training. What happens to that data once it reaches a third-party provider. Whether a person reviews what the model produces. And somewhere in it, a question the certificate simply doesn’t speak to: how do you govern all of this?
It’s a moment a lot of well-run companies are meeting for the first time, and the certificate they expected to rely on wasn’t designed to answer it.
An AI section in security reviews
- Asks
- Which models, where data goes, who reviews output
- Raised by
- Enterprise and regulated buyers
Information security
- Covers
- Risk, controls, suppliers, incidents
- Built for
- Managing information security well
Governing AI responsibly
- Missing
- Model intent, training data, human oversight
- Answered by
- An AI management system
Why ISO 27001 doesn’t cover your AI
ISO 27001 is a standard for managing information security, and it does that job well. It asks you to identify your risks, decide how to treat them, and show that you operate the controls you claim to. Its Annex A covers access, supplier relationships, change management, incident response, and the other security fundamentals.
What it doesn’t do is address artificial intelligence specifically. It treats an AI system much as it treats any other asset or supplier. That isn’t a shortcoming - it’s simply the scope the standard was written for. The questions a buyer now asks about AI, such as where your training data comes from, whether a model’s output can be explained, how you catch it when it gets something wrong, and what the system is and isn’t meant to be used for, sit outside that scope.
The certificate tells a customer you manage information security. It doesn’t tell them whether you manage AI responsibly. Those have become two different questions.
What buyers are actually asking
The AI section of a security review tends to return to the same concerns, phrased in various ways. None of it is unreasonable. These are the AI versions of questions buyers have asked about data and security for years. The challenge is that an existing evidence pack mostly doesn’t answer them.
Where your ISMS already reaches: and where it stops
This is a point that often gets lost. If you already run an ISO 27001 ISMS, you are not starting from zero, and a good deal of the groundwork carries over. A company with a working ISMS can answer a fair share of the AI questionnaire today, by pointing controls it already has at a new kind of asset.
That second layer is the genuine gap, and it’s where ISO 42001 comes in.
Where ISO 42001 fits
ISO 42001, published in 2023, is the management-system standard for artificial intelligence. It sits alongside ISO 27001 and works in a familiar way - governance, risk, defined controls, evidence - but focused on the AI lifecycle rather than information security in general. It covers much of what the questionnaire now asks about: AI risk and impact assessment, data governance for AI, human oversight, and accountability for how a system behaves.
For a company already certified to ISO 27001, taking it on is less daunting than it first appears, because the management-system machinery is already running. You would be extending something that works rather than standing up a second system from scratch.
Not every company needs an ISO 42001 certificate today, and the regulatory picture, including the EU AI Act, which we looked at in a separate piece, is still settling. For many SaaS companies the sensible step right now is to close the clear gaps and be able to answer the questionnaire credibly, with a route to formal certification when the risk, or a customer, genuinely calls for it.
What to do about it now
Roughly in order of effort:
Begin with an AI inventory. You can’t govern what you haven’t listed, and most companies have more AI in their product and operations than they expect.
Assess your AI providers as the suppliers they are, and check what your contracts say about training and data use.
Put a short, workable acceptable-use policy in place for staff.
Map where customer data actually flows when an AI feature touches it.
Map the questionnaire itself against what your ISMS already answers, so you know where the real gaps are rather than guessing.
That much will usually move you from “we can’t answer this” to a considered, evidenced response, which is often what the deal actually needs.
How Stenward helps
This is close to the work we do, so we can be practical about it. We tend to start by mapping the AI questions you’re being asked against the ISMS you already have, so it’s clear what’s genuinely covered and what still needs work.
From there we help extend the system to close the real gaps - the AI inventory, supplier assessments, data-flow mapping, and the acceptable-use and oversight policies - building on the ISO 27001 foundation you already have rather than duplicating it. Where ISO 42001 is the right next step, we help you work towards it; where it isn’t needed yet, we’ll say so and help you answer the review well in the meantime. The work is led by someone senior who has seen these questionnaires from both sides.
If that’s where you find yourself at the moment, we’d be glad to talk it through.
The goal is simply for the AI section of a security review to become something you can answer with confidence, rather than something that holds up the contract.
Book a call