A business has published an AI policy and asked every employee to acknowledge it. Six months later, a department adopts a new assistant without assessment, a supplier changes its service terms and staff begin using outputs for a more important decision. The policy remains unchanged while the actual exposure grows.
ISO 42001 auditing should examine whether governance follows these changes. For Canadian and American organizations, the useful question is whether decisions about AI use are controlled and supported by evidence. A document review is necessary, but it cannot establish how the organization manages applications in practice.
An internal AI management system audit evaluates conformity and effectiveness within an agreed scope. It differs from testing the technical performance of one model, although technical evidence may be relevant to the management system assessment.
Decide what the audit must establish. Is the organization controlling newly approved applications? Are significant changes reassessed? Can employees report an incident? Do risk treatment decisions correspond to controls that operate?
Record the scope, criteria, methods and competence needed. ISO 19011:2026 provides auditing guidance, while the applicable management system requirements and organizational commitments establish the assessment criteria.
Follow ISO 42001 Auditing application through its lifecycle
Choose a representative AI use and trace the decisions surrounding it. Begin with the business request, then examine approval, assessment, supplier review, deployment, user training, monitoring and changes.
For an illustrative recruitment assistant, relevant evidence might include the approved purpose, identified affected groups, evaluation criteria, human review arrangements and records of challenges or errors. The auditor should understand how the tool influences decisions before selecting evidence.
A lifecycle trail often reveals mismatches between departments. Procurement may approve the supplier while nobody approves the intended use. Training may explain general principles while omitting the restrictions associated with the application.
Select samples according to exposure
An audit programme should consider factors such as potential consequences, changes, previous findings and operational performance. A high-impact application with unresolved concerns deserves different attention from a limited internal drafting tool.
Within an application, select evidence across relevant periods, user groups or operating conditions. Include changed uses and exceptions where these help evaluate control effectiveness. Record why the samples were selected.
ISO 42001 auditing uses sampling to support an evidence-based conclusion, not a promise that every output is correct. Describe access restrictions and evidence limitations, especially where confidential data or proprietary supplier information affects the work.
Test the boundary between approval and actual use
Compare the approved purpose with observed activity. A tool approved to summarize public information may later receive internal customer records. A system approved to support a reviewer may gradually become the default decision maker.
Ask employees to demonstrate their normal workflow. Examine configuration, access permissions and relevant logs where appropriate and authorized. Compare those observations with the inventory, assessment and user guidance.
Do not infer misconduct from an inconsistency alone. Establish the facts and determine whether the issue reflects unclear instructions, insufficient access controls, uncontrolled change or another supported cause requiring further investigation.
Examine human oversight as a control
A policy may require review without explaining what the reviewer must verify. The auditor should examine competence, available information, authority and evidence that review occurs at the intended point.
Take a sample of outputs and examine the associated review records where retention is appropriate. Ask how reviewers recognize unreliable results, resolve uncertainty and stop inappropriate use. Consider whether workload makes the control unrealistic.
An approval button is weak evidence if the reviewer cannot assess the relevant risk. The audit should evaluate the design and operation of the control within the organization’s defined requirements, rather than assume a recorded click proves meaningful oversight.
Connect risk treatment to implementation
Review the relationship between identified risks, impact assessment, necessary controls and the Statement of Applicability. Check whether documented treatment decisions have owners and whether implementation can be demonstrated.
For example, a risk assessment may identify inaccurate customer communication. The treatment could include restricted purposes, required source checks and escalation for uncertain cases. Audit evidence should establish whether those controls are present in the actual workflow.
Where the organization relies on a supplier, examine how responsibilities are allocated. A supplier assurance statement may be relevant evidence, but the organization still needs to understand its own obligations and the limits of the assurance received.
Include changes and incidents
Select a significant update or reported incident and follow the response. Was the concern captured? Did someone assess the consequences? Were affected activities restricted when necessary? Was the assessment updated before normal use resumed?
An incident can reveal how governance behaves under pressure. Compare the written reporting route with what employees actually did. Consider whether the organization learned from the event and assessed similar uses elsewhere.
Where no incidents have been reported, do not automatically conclude that reporting works. Interview users and examine whether concerns could be recognized and raised without unnecessary barriers.
Report findings with proportion and clarity
A finding should identify a requirement, objective evidence and the demonstrated gap. Avoid broad labels such as “unsafe AI” unless the audit’s evidence and criteria support that conclusion.
Separate confirmed nonconformities from questions requiring further assessment and improvement opportunities. Explain material limitations. A management system audit is not universal legal assurance, a cybersecurity penetration test or certification of each individual model.
Kadmar Consultants provides AI governance audit support and training. A useful engagement begins with agreed boundaries, competent evaluation and a reporting approach that management can use to assign corrective actions.
Preserve a usable evidence trail
Agree how evidence references will identify the application, version, date and relevant decision. Record who authorized access to each evidence source. This helps the organization investigate a finding without distributing unnecessary sensitive information. When a supplier controls important records, establish access arrangements early. If access is unavailable, assess the consequence for the audit objective and report the limitation clearly. An evidence index should make findings traceable while respecting the organization’s information handling rules.
Frequently asked questions
Can our existing quality auditors perform this work?
Their audit skills may provide a foundation. They also need competence appropriate to AI governance, relevant uses and the audit scope. A multidisciplinary team or technical specialist may be needed for complex applications.
Does an audit require access to source code?
Not in every case. Evidence needs depend on the organization’s role, scope and audit objectives. The auditor must determine whether available information supports a conclusion and disclose restrictions that materially limit the assessment.
How should confidential information be handled?
Agree access, confidentiality, evidence recording and retention arrangements before the audit. Use authorized demonstrations or appropriately protected records where suitable. Avoid copying sensitive inputs into an audit report when a more controlled evidence reference is sufficient.
Is this the same as a certification audit?
No. Internal assurance and certification have different roles. A certification body independently evaluates the management system for certification. Internal audit support helps the organization evaluate its own controls and address weaknesses.
Make the audit useful to decision makers
Ask Kadmar Consultants to help define an audit trail around an application that matters to your organization. Well-scoped ISO 42001 auditing should clarify whether ownership, assessment and oversight remain effective as AI use changes.


