Series · ISO/IEC 42001

Preparing for an AIMS

A twelve part series on what to do before anyone mentions policies, controls or certification. Published on LinkedIn, collected here, and gathered into a single document when the series finishes.

Parts published  9 of 12 Standard  ISO/IEC 42001:2023

The series so far

A 107 second silent summary of parts 1 to 9. Everything in it is written out below.
  • Part 1

    Do you know what AI you are already using in your organisation?

    There is the AI purchased on purpose, the AI that arrived as a new feature on an existing platform, and the AI individuals brought in themselves. An inventory is the first step, and the one step that delivers value even if nothing else follows. You cannot govern what you cannot see.

    Read part 1 on LinkedIn
  • Part 2

    Understanding your organisational context

    Clause 4.1 asks why AI matters to this organisation specifically: whether you develop it, provide it, use it or all three, what regulators and customers expect, and what you are actually trying to achieve. An AIMS designed without context is just a template, and a template will not survive an audit or a real incident.

    Read part 2 on LinkedIn
  • Part 3

    Who cares how you use AI?

    Most organisations list customers, regulators and shareholders, then move on. The AI version of the exercise reaches further: the people your AI makes decisions about, the staff expected to override it, your suppliers’ suppliers, and the sector regulators already publishing AI expectations. This is where the reason for each control comes from.

    Read part 3 on LinkedIn
  • Part 4

    What does your AIMS actually cover?

    Scope too broad and you drown in documentation. Too narrow and you have certified part of the picture while the exposure sits outside it. Exclusions need a real justification: a vendor-managed tool with no personal data and no decision impact holds up, a gut feel decision doesn’t.

    Read part 4 on LinkedIn
  • Part 5

    Leadership and the AI policy

    Clause 5 asks top management to demonstrate commitment rather than sponsor it or approve it: a policy leadership actually shaped rather than signed, a named senior owner whose objectives include it, real resources rather than finding time alongside the day job, and roles defined down to who can approve a new AI use and who can prevent one. Two pages is plenty. The test is whether a new starter could read it and know whether pasting a client contract into a public chatbot is acceptable here. If not, it is a values statement rather than a policy.

    Read part 5 on LinkedIn
  • Part 6

    Why one assessment is not enough when assessing the risks of AI

    Clause 6.1 asks for two things organisations routinely treat as one. The AI risk assessment covers risk to the organisation: model drift, biased outputs, supplier dependency, data leakage. The impact assessment covers the effect on individuals, groups and society: loss of access to a service, unfair treatment, loss of privacy. Different questions, and different people. If you already run DPIAs the second will feel familiar, but do not assume a DPIA covers it, because fairness, safety and autonomy sit outside data protection. Assess per system, use one consistent scale, and record the reasoning, because that is what an auditor, or a regulator after an incident, will want to see.

    Read part 6 on LinkedIn
  • Part 7

    Risks when using AI: treating them and setting objectives

    Reduce, accept, avoid or transfer, plus a fifth that AI adds in practice: constrain. A human in the loop for high stakes decisions, advisory rather than automated use, or limits on what data the model can see. Annex A is the catalogue of controls and the Statement of Applicability records what you adopted, what you did not, and why. Then Clause 6.2, where good intentions become measurable: objectives with an owner, a measure, a deadline and a review point, rather than just a line in your policy about using AI responsibly.

    Read part 7 on LinkedIn
  • Part 8

    Do you have what it takes to support and maintain your AI management system?

    The unglamorous clause, and the one where most audits find the gaps. Resources, because governance is work and someone needs the time. Competence, covering not just the data scientist but the procurement lead evaluating a vendor, the manager approving a use case and the caseworker expected to challenge a recommendation. Awareness that is role relevant, repeated and evidenced. Communication with staff, customers and regulators. And the records themselves, controlled, versioned and findable. Start the evidence trail now: a dated log of decisions and who made them is worth more than a polished procedure written the week before an audit.

    Read part 8 on LinkedIn
  • Part 9

    Governing the AI lifecycle

    Clause 8 and Annex A are where the management system meets the actual systems. The lifecycle framing is useful even if you never certify: inception and who approved the objective, design and data choices, verification that it does what was intended, deployment with release criteria and a rollback plan, monitoring for drift and incidents, and retirement, including what happens to the people relying on it. Data provenance matters more than most organisations appreciate. And if you buy rather than build, your supplier’s AI is your risk. Most organisations already have change control, supplier management and testing. The ask is that AI systems pass through them rather than around them.

    Read part 9 on LinkedIn

Still to come

Three further parts, starting with how you will know whether any of it is working. This page is updated as each one is published.

When the twelve parts are complete, the whole series will be available here as a single document to download and work through.

Get in touch