Auditing AI systems
ISO 19011:2026 clause 7 · ISO/IEC 42006:2025 clause 7.1 — Competence and evaluation of auditors · Competence of personnel
Nothing in the audit standards says an AI engagement needs an AI specialist. What ISO 19011:2026 clause 7 requires is that the function determine the competence an engagement needs, set criteria for evaluating it, and evaluate people against them. For AI audits that determination is usually the step nobody has taken.
§ 1 — Who it binds
Which instrument reaches whom
Internal audit functionCertification body
ISO 19011 is guidance, not a requirement, and applies to any organisation that needs to plan and conduct audits of management systems or manage an audit programme. ISO/IEC 42006 is different in kind: it states requirements, and it binds bodies providing audit and certification of AI management systems rather than internal functions. Most readers of this page sit in the first population and are looking at the second for its answer.
§ 2 — In practice
Competence is determined, then evaluated
Clause 7 of ISO 19011:2026 treats competence as something a function works out, not something a person arrives holding. Its sequence is explicit: 7.2 determining auditor competence, 7.3 establishing the auditor evaluation criteria, 7.5 conducting the auditor evaluation. Applied to an AI engagement, 7.2 is the step that gets skipped. A function that has not written down what an auditor must be able to do — read a model card sceptically, tell a validation split from a deployment population, judge whether a threshold was set or inherited — has no criteria to evaluate anyone against at 7.3, and its evaluation at 7.5 collapses into a view about a person. The pattern is the same one the engagement itself runs into when it asks what the system should be measured against; see evaluation criteria for an AI audit.
The conformity-assessment side of the field met the same question and answered it in writing. ISO/IEC 42006:2025 — the requirements for bodies providing audit and certification of artificial intelligence management systems — devotes 7.1 to competence of personnel, separates external auditors and technical experts into 7.3 of the same clause, and requires personnel records at 7.4. A body cannot satisfy those by asserting that its auditors are experienced. What is worth noticing is structural rather than legal: when the question of who may audit an AI management system had to be settled for certification bodies, the answer took the form of a standard of its own. An internal function is under no obligation to follow that standard, and is answering the same question.
The practical answer is usually a team rather than a person, and both instruments accommodate that — technical experts are named alongside auditors precisely because the two are different roles. The distinction that matters in the file is between an expert who is engaged and an expert who is relied on: where the conclusion rests on the expert’s judgement, their independence from the activity under audit becomes a live question and should be declared rather than assumed. The second distinction is between AI awareness and AI method. A two-hour module on generative models is awareness. What an engagement consumes is method — how a population is defined for a system that has no stable one, how a control is tested when its output is a distribution — which is the subject of testing a probabilistic control.
§ 3 — What a weak answer looks like
The training log as a competence file
A training log offered as a competence file. Everyone on the team completed the same introductory module, the completion percentage is reported to the audit committee, and no document anywhere states what this engagement actually required of the people staffed on it. The tell is that the log would look identical if the work had been a payroll review. Clause 7 asks a question about the work first and the people second, and a log answers it in the wrong order.
§ 4 — What discharges it
What a competence file has to show
The artefacts an assessor asks to see, and what makes each one sufficient rather than merely present.
01
The competence determination for AI engagements, made before staffing
What 7.2 asks for: the knowledge and skills this kind of engagement needs, stated as things an auditor must be able to do rather than as tools or technologies they should have heard of.
02
Evaluation records under 7.5, one per auditor on the engagement
Each person assessed against the criteria set at 7.3, with the method used and the outcome. A record showing everyone met every criterion, produced after the team was chosen, is a ratification rather than an evaluation.
03
The technical-expert engagement record, with an independence declaration
Who was brought in, for which question, what they were given, and what part of the conclusion rests on them. Where the expert has worked on the system under review, the declaration is what lets a reader judge the reliance.
04
Continuing development in AI method rather than AI awareness
Sampling and population definition for model outputs, evaluation design, test-environment control. Attendance at an introductory session on generative models evidences interest, and answers nothing clause 7 asks.
§ 5 — Worked example
Worked example — the seconded data scientist
A financial institution’s internal audit function schedules a review of the monitoring controls over its credit model. The team is two experienced auditors who have each completed the firm’s AI foundation course, supported by a data scientist seconded from the model risk function for four days. The engagement file records the secondment in the resource plan.
Is the competence requirement discharged?
Two gaps, and neither is about the people. The first is that no document states what the engagement required. Without a determination under 7.2 there are no criteria at 7.3, so the resource plan records who was available rather than who was needed, and nobody can say afterwards whether the team was right. The second is the secondment. A data scientist from model risk is very likely the most capable person in the building on this system, and is drawn from a function whose work forms part of what is under review. That does not disqualify the reliance; it makes the reliance something that has to be declared, bounded and recorded — which question was put to the expert, what was accepted on their judgement, and what the auditors established themselves. Neither gap is expensive to close, and both stay invisible until someone challenges the conclusion.
§ 6 — Elsewhere
The same question, asked on the other side
Where another instrument addresses the same obligation. These are correspondences, not comparisons — the Council does not rank one framework against another.
Article 26(2) requires a deployer to assign human oversight to people with the necessary competence, training and authority. It puts a competence question on the operating side of a system; clause 7 puts one on the side reviewing it.
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.
§ 7 — When it applies
Current editions, and why the date is not a deadline
27 May 2026
BSI publication of ISO 19011:2026, the current edition. An edition date for voluntary guidance, not a compliance deadline — and a reason to re-check clause numbers, since material written against the 2018 edition may now be misnumbered.
31 July 2025
BSI adoption of ISO/IEC 42006:2025 as BS ISO/IEC 42006:2025, identical to the ISO text. Again an edition date rather than a deadline; ISO’s own publication date is not asserted here.
§ 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.
- ISO 19011:2026, Guidelines for auditing management systems — clause 7 Competence and evaluation of auditors, sub-clauses 7.2, 7.3 and 7.5. Current edition published by BSI on 27 May 2026.
- ISO/IEC 42006:2025, Requirements for bodies providing audit and certification of artificial intelligence management systems — clause 7.1 Competence of personnel, with 7.3 and 7.4.
- Edition dates: ISO 19011:2026 published by BSI 27 May 2026; BS ISO/IEC 42006:2025 published 31 July 2025.
§ 9 — Also read
Also read
Testing a probabilistic control
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 free AI Literacy Certificate is open to any adult today, and the register lists what is open for enrolment.