AI Governance Audit

The buyer problem

Most organizations that build an AI governance program never have it independently tested. A written policy, an assigned owner, and a risk register are necessary but not sufficient: they describe intent, not verified practice. The gap between a documented program and one that actually catches problems is exactly what an internal audit function or an outside auditor is built to find, and it's the same gap a board or regulator will ask about first. The practical difficulty for a buyer is that 'AI audit' means different things to different providers: some offer a compliance-checklist review, others a technical model-testing engagement, and few clearly state which named framework their methodology is actually built on.

What a ai governance audit engagement covers

An AI governance audit engagement independently assesses whether an organization's AI governance program - its policies, roles, inventory, risk assessment process, and controls - is both documented and operating as designed. A credible engagement is explicitly scoped against one or more named frameworks (most commonly the NIST AI RMF, ISO/IEC 42001, or the IIA's AI Auditing Framework) rather than a generic, unsourced checklist, and produces a findings report with specific, closable gaps rather than a pass/fail score.

Methods and techniques

• Governance-structure review: confirming a named accountable owner, a documented AI inventory, and defined escalation paths exist and are followed in practice, not just on paper
• Framework gap assessment: mapping current-state controls against a named framework's specific functions or clauses (e.g. NIST AI RMF's Govern/Map/Measure/Manage, or ISO 42001's Annex A controls) and identifying specific unmet requirements
• Control testing: sampling actual AI system records (risk assessments, model validation results, monitoring logs) to confirm documented controls were genuinely performed, not just described in policy
• Three Lines Model application: for internal-audit-led engagements, explicitly separating first-line (build/operate), second-line (risk/compliance oversight), and third-line (independent audit) responsibilities per the IIA's model
• Findings and remediation tracking: a written report of specific, closable gaps tied to the framework clause or control they violate, not a vague maturity score

What to verify before you retain

• Named framework basis. Ask which specific framework(s) the audit methodology is built on (NIST AI RMF, ISO 42001, IIA's AI Auditing Framework) and request a sample findings report or methodology document, not just a marketing description.
• Auditor independence. If the same firm designed your governance program, confirm who is actually performing the audit and whether there is a genuine independence boundary, consistent with the IIA's Three Lines Model.
• Scope of technical testing. Clarify whether the engagement includes actual technical testing of AI systems (model validation, red-teaming) or is a documentation/process review only - these are very different deliverables often marketed under the same 'AI audit' label.
• Credential and methodology transparency. For internal-audit-adjacent engagements, ask whether the lead auditor holds a relevant credential (e.g. CIA) and whether the firm can show prior AI-specific audit work, not just general IT audit experience.

Questions to put in your RFP

1. Which named framework(s) does your AI governance audit methodology map to, and can you share a redacted sample findings report?
2. Is this a documentation review, a technical testing engagement, or both? What specifically is out of scope?
3. Who performs the audit, and what is their relevant credential or prior AI-audit experience?
4. How do you handle independence if your firm was also involved in designing our governance program?
5. What does a typical findings report look like, and how are findings tracked to closure?
6. Can you audit against more than one framework in a single engagement (e.g. NIST AI RMF and ISO 42001 jointly)?

Red flags

• A firm that claims its audit 'guarantees compliance' with any framework - no independent audit can guarantee anything; it can only report findings as of the engagement date.
• No named framework basis at all - a generic, unsourced 'AI maturity checklist' is not the same as a framework-grounded audit.
• Refusal to clarify whether the engagement is documentation review only versus technical testing - this ambiguity is a common source of buyer disappointment after the engagement starts.
• A firm that both built your governance program and offers to 'independently audit' it with no disclosed independence safeguard.

Frameworks referenced

Named frameworks relevant to this category. Listed for context; they do not endorse this index or any vendor. Verify any framework alignment claim directly against the issuing body.

NIST AI RMF

Artificial Intelligence Risk Management Framework (AI RMF 1.0). The NIST AI RMF provides voluntary guidance to help organizations that design, develop, deploy, or use AI systems manage risks to individuals, organizations, and society. It is intended for use across sectors, use cases, and organization sizes, and is not tied to any specific technology. The framework organizes AI risk management around four functions, Govern, Map, Measure, and Manage, oriented toward seven characteristics of trustworthy AI: valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. Read more →

ISO 42001

ISO/IEC 42001:2023 - Information technology - Artificial intelligence - Management system. ISO/IEC 42001:2023 specifies requirements, with guidance, for establishing, implementing, maintaining, and continually improving an AI management system (AIMS) within an organization. It is the first international management-system standard written specifically for AI and is designed for any organization, of any size or sector, that provides or uses AI-based products or services. It addresses AI-specific concerns such as algorithmic bias, transparency, and lifecycle impacts rather than treating AI as generic IT. Read more →

IIA AI Auditing Framework

The IIA's Artificial Intelligence Auditing Framework. The IIA's AI Auditing Framework gives internal auditors principles-based guidance for assessing an organization's AI governance, risk management, and controls. It covers AI risk from strategy and governance through model development, deployment, and monitoring, including generative AI and large language models. It is intended for internal audit functions, chief audit executives, and audit committees performing assurance or advisory engagements on AI use, and it is explicitly structured around the IIA's Three Lines Model. Read more →

Three Lines Model

The IIA's Three Lines Model. The IIA's Three Lines Model is a governance and risk-management model that assigns distinct roles to a governing body, management (split into first-line and second-line roles), and an independent internal audit function, the third line. Internal audit practitioners and the IIA apply this same structure to AI governance: first-line roles build, deploy, and operate AI systems and own the resulting risk; second-line roles provide AI-specific risk expertise, monitoring, and challenge; and internal audit provides independent assurance that AI risk is being governed and managed effectively. It is a principles-based model meant to be adapted to an organization's structure rather than a prescriptive control checklist. Read more →