Control · C-063 · Monitoring, Assurance & Incidents

Independent assurance and internal audit of AI

Second-line review and internal audit cover AI governance and the highest-tier systems on a risk-based plan, test systems as well as documents, and track their findings to closure.

Control type: Detective
Family: 11 Monitoring, Assurance & Incidents
Risks addressed: 6
Audience: Board · Heads-of
Frameworks: Clause 9 · GOVERN 4 · Principle 4
Responsible AI domain: 11 Monitoring, Assurance & Incidents
Lifecycle stage: 7 Operate & monitor
Reviewed: 2026-11-10

What it does

It gives the board an independent view of whether AI controls exist and work. The second line reviews and challenges the first on a planned basis, and internal audit provides an opinion that does not depend on either. The head of internal audit owns the audit plan and the chief risk officer owns second-line review. Both operate on a cycle set by risk, with coverage of AI governance at least annually.

How to implement it

Add AI to the audit universe and the second-line review plan.
Cover the governance framework and a sample of high-tier systems.
Test the operation of controls on live systems, and not only the existence of policy.
Test the completeness of the inventory and the accuracy of tiering.
Build or bring in the skills needed to challenge technical work.
Report findings to the audit committee and track actions.

Evidence it operates

Design: audit and review plans that include AI; a statement of the skills available. Operation: reports with findings and ratings; action tracking showing closure; evidence of testing on live systems; reporting to the audit committee.

Where it fits

Operates over the whole framework in the third line, with the second line beneath it. It is the primary control for weak independent assurance of AI and supports misclassified AI risk tier by sampling classifications. It is distinct from independent model validation, which examines individual models.

Limits

Assurance samples, so it can miss what it did not select. Auditors without relevant skills will check documents and leave the system untested. Findings arrive after the fact and change nothing until they are acted on. An audit function that reports no findings on a new and fast-moving risk should be questioned as closely as one that reports many.

Risks this mitigates

Primary mappings

Weak independent assurance of AI Medium: Places AI on the second-line and internal audit plans with tracked findings. Second-line review and internal audit do not cover AI with sufficient skill or frequency, so control failures persist undetected.

Supporting mappings

Inadequate independent model validation High: Checks that validation happens and is independent. Validation is absent, is carried out by the model's own developers, or is not proportionate to the model's tier, so weaknesses are discovered in production.
Mis-specified fairness measurement Medium: Assurance checks that the stated approach was followed. Fairness is tested with a metric, threshold or grouping that does not match the legal test or the harm, so a system passes while remaining unfair.
Misclassified AI risk tier High: Samples classifications for accuracy. A system is assigned a lower risk tier or regulatory class than it warrants, so it receives lighter controls than the harm it can cause.
Overstated AI claims Medium: Tests stated governance against practice. The organisation misdescribes what its AI does, how well it performs or how it is governed, to customers, investors or regulators.
Untracked regulatory change Medium: Tests that the register is complete and acted on. New or amended AI obligations, including changed application dates, are not identified and translated into controls in time.

Framework and clause references

Framework	Clause	Title
ISO/IEC 42001:2023	Clause 9	Performance evaluation Internal audit.
NIST AI Risk Management Framework (AI RMF 1.0)	GOVERN 4	A culture that considers and communicates AI risk Challenge.
SS1/23: Model Risk Management Principles for Banks	Principle 4	Independent model validation Independent review.

Related terms

How to cite

Axiom Verity (2026). Independent assurance and internal audit of AI. Reference Library, C-063. https://axiom-verity.com/controls/internal-audit-of-ai. Reviewed 2026-11-10.

Found an error or a better source? Report a correction.