Skip to content
NEWWhich kind of AI user are you?
Take the two-minute check
AIAC AI ASSURANCE COUNCIL™

ISO/IEC 42001

Clause 9.2, with 9.2.1 and 9.2.2 — Internal audit · General · Internal audit programme

Clause 9.2 is not flat. The standard’s own contents page subdivides it into 9.2.1 General and 9.2.2 Internal audit programme, and the split carries weight: the first governs what an individual audit determines, the second the programme deciding which audits happen, how often, performed by whom, and how their results reach management.

§ 1 — When it applies

Which edition, and why no deadline attaches

  1. 31 December 2023

    Publication date of the BSI adoption of the first edition; an edition date, not a compliance deadline.

  2. 31 March 2026

    European and UK adoption, BS EN ISO/IEC 42001:2026 — BSI states no technical differences in the requirements or guidance.

§ 2 — In practice

Two sub-clauses, and the one commentary loses

Take the structure first, because a good deal of commentary has it wrong. The contents page of ISO/IEC 42001:2023 lists 9.2.1 General and 9.2.2 Internal audit programme beneath 9.2 Internal audit, and the NIST-hosted crosswalk names the same two sub-clauses independently. Anything written against a flat 9.2 is citing a level of the standard that does not exist, and the practical cost is that the programme duty disappears — it lives entirely in 9.2.2, and a page treating the clause as a single requirement to "audit the AIMS" reduces it to whether an audit happened. It is worth saying plainly that this correction runs against a note the Council itself previously held: absence of a sub-clause in a partial listing is not evidence that the sub-clause does not exist.

An internal audit of an AI management system is not a model validation, and functions staffed by data scientists reliably produce the second when asked for the first. The distinction is not about rigour. A validation asks whether a model performs; 9.2.1 asks whether the management system conforms to the organisation’s own requirements for it and to the standard, and whether it is effectively implemented and maintained. Those questions are answered by looking at whether the impact assessment cadence was honoured, whether risk owners exist, whether the applicability statement was revisited — none of which requires opening a model at all. Conducting an AI audit sets out the engagement mechanics; what the clause adds is that the subject is the system of management rather than the system being managed.

Objectivity and impartiality is where small AI functions come apart, and it is a selection duty rather than a character judgement. The person who wrote the AI policy cannot audit the AI policy, however scrupulous they are, because the clause is about the arrangement and not about them. In an eleven-person function that looks impossible until it is written down, at which point the ordinary answers appear: an auditor drawn from another assurance function, a peer from a different business unit, a contracted auditor, or a reciprocal split in which each person audits what the other owns. Competence has to be set alongside independence — what an auditor of an AIMS must be able to do is its own determination, which is the subject of auditor competence for AI audits.

The programme is a judgement, and a flat annual round-robin does not embody one. 9.2.2 asks for frequency, methods, responsibilities, planning requirements and reporting to be planned and maintained as a programme, taking account of the importance of the processes concerned and the results of previous audits. Both of those inputs are testable. Importance means some processes are audited more often than others and the file says why; previous results mean this cycle differs from the last one in a way that can be traced to what the last one found. A schedule covering every clause once a year, unchanged from the year before, satisfies neither and is visibly a calendar rather than a programme — a distinction the audit programme for AI systems works through from the audit side.

Nothing happens to an organisation that ignores ISO/IEC 42001. It is a voluntary standard, no regulator enforces it and no fine attaches to any part of it, and pages implying otherwise are describing something else. The real consequence of a hollow internal audit is narrower and quite concrete: a certification body operating to ISO/IEC 42006 examines internal audit as part of what it assesses, and a certificate can be refused, suspended or withdrawn. An internal audit is also not a certification audit and cannot be substituted for one in either direction — the external body’s work does not discharge 9.2, and the organisation’s own audit does not shorten the body’s.

§ 3 — Who it binds

Whose audit, and how often

Audited organisation

Every organisation operating an AI management system to ISO/IEC 42001, recurrently rather than once. Clause 9.2 binds the organisation and not any external party: an audit performed by a certification body, a customer or a consultant is not an internal audit and does not discharge it. The programme duty applies even where a single audit would cover the whole AIMS, because the programme is what shows why that coverage was the right one.

§ 4 — What discharges it

What the audit file must show

The artefacts an assessor asks to see, and what makes each one sufficient rather than merely present.

  1. 01

    The internal audit programme, with its risk basis stated

    Frequency per process and the reasoning behind each choice, so that a reader can see which parts of the AIMS the organisation considers to matter more. A programme with one frequency for everything has recorded a calendar and no judgement.

  2. 02

    Auditor independence declarations, one per engagement

    For each auditor, what they did not audit and why — the policy they drafted, the register they own, the process they run. A blanket declaration of independence at function level cannot address the conflicts that actually arise in a small team.

  3. 03

    Audit reports with findings graded to a stated methodology

    A grading scale defined before the work, applied consistently, and visible in the report. Where every finding is minor and no scale is published, the grading is a negotiation outcome rather than an assessment.

  4. 04

    Competence criteria for auditing an AIMS, and evidence against them

    What an auditor of this management system must be able to do, written before staffing, with per-person evidence of education, training or experience against it. Attendance at an introductory session evidences attendance.

  5. 05

    The previous-results input, showing the programme moved

    What the last cycle found and what changed in this cycle because of it — a process brought forward, a method altered, coverage widened. Where nothing changed, the file should record that this was considered and why it was right.

  6. 06

    The record of reporting to relevant management, dated

    Who received the results, when, and in what form, including where the recipient was not the person who commissioned the work. Results circulated only within the function audited have not left the boundary they were meant to cross.

§ 5 — Worked example

Worked example — the auditor who wrote the policy

A software company with an eleven-person AI function schedules its AIMS internal audit. It is performed by the AI governance lead, who wrote the AI policy, maintains the risk register and owns the applicability statement. The programme is a single annual audit covering every clause in one two-day pass, scheduled for the month before the certification visit. The report raises two minor observations, both closed before the certification body arrives.

What will the certification body ask about first?

The auditor, and not because of competence. Somebody who wrote the policy, maintains the register and owns the applicability statement is reporting on their own work, and the selection duty is not satisfied by that person being conscientious. Small functions hit this constantly and the ordinary answers exist, but none is available if nobody has written down that the problem is there. The programme is the second question. One annual pass over every clause is a schedule: it embeds no judgement about which processes matter more, and it cannot show that the previous round changed anything, because the previous round covered everything equally too. The timing is the third and the most telling. An audit run the month before the external visit, with its findings closed inside the same window, tests readiness for an inspection rather than conformity of a management system — and a report in which nothing was inconvenient enough to argue about is the signature of exactly that.

§ 6 — What a weak answer looks like

A dress rehearsal is not an audit

The certification rehearsal offered as an internal audit. It is scheduled to land just before the external visit, it walks the clauses in numerical order, and everything raised is drafted so it can be closed before the assessor arrives. That is useful preparation and it is not what the clause asks for, because it tests readiness rather than conformity and structurally cannot produce an inconvenient result. The signature is a report containing nothing anybody had to argue about, in a management system that has not yet been operated for a full cycle without an audience.

§ 7 — Elsewhere

The audit profession’s side of the same duty

Where another instrument addresses the same obligation. These are correspondences, not comparisons — the Council does not rank one framework against another.

  • Auditing AI systems

    ISO 19011:2026 addresses the establishment and management of an audit programme. It reaches the same territory as 9.2.2 from the audit profession’s side, as guidance rather than as a requirement on the organisation.

  • Auditing AI systems

    Competence for auditing a management system is determined, then evaluated against criteria set in advance. Clause 9.2.2 asks the organisation to select auditors for objectivity and impartiality, and competence is the other half of that selection.

  • Auditing AI systems

    ISO/IEC 42006:2025 governs the bodies certifying AI management systems. An internal audit under 9.2 and a certification audit under 42006 are different exercises performed by different parties, and neither discharges the other.

A correspondence indicates that two instruments address the same underlying obligation. It is not a mapping endorsed by either body, not a statement that one satisfies the other, and not a judgement about which is more demanding.

§ 8 — provenance

The provision itself

This page sets out what the instrument requires and what discharges it. The official text is the authority — these go straight to it.

§ 9 — Also read

Also read

Certification

Assessed on the same standard of evidence

Every Council credential is examined on applied judgement against a published anchor, set out the way the obligations on this page are. The Certified Applied AI Associate, the Council’s flagship, is open to any adult today, its fee waived by a coupon code from the AI-Readiness Check, and the register lists what is open for enrolment.

Flagship · AIAC-CAAACertified Applied AI Associate30 minutes · 18 or over · on the register
Open now →The credentials open for enrolmentThe register →