AI Governance and Risk · Topic 2 of 6
AI Governance and Program Management
Auditing AI governance: ownership and accountability, AI policy, AI inventory, committees, three lines, awareness and AI literacy, and workforce impact.
- 5 min
- Every AI system needs a named, accountable business owner.
- The AI inventory is the audit universe; test it for completeness against other sources.
- A policy is design; signed approvals and minutes show operation.
What it is
AI governance is the system of direction and oversight that makes sure AI supports organizational goals, stays within risk appetite and complies with law and policy. An AI program is how that governance is run day to day: the inventory, policies, roles, approvals, training and reporting.
The auditor evaluates whether governance is designed well (the right roles, policies and approvals exist) and operating (people actually follow them). The commonest finding is not a missing policy but a policy nobody can show was applied.
How it works
Ownership and accountability
The outline asks the auditor to evaluate whether the organization has defined ownership of AI risk, controls, procedures, decisions and standards. Every AI system should have a named business owner who is accountable for its outcomes.
| Role | Typical responsibility |
|---|---|
| Board or board committee | Sets AI strategy and risk appetite, receives reporting |
| Senior management or AI steering committee | Approves policy, prioritises use cases, resolves conflicts |
| AI system owner (business) | Accountable for the use, its outcomes and accepting residual risk |
| Data science or development team | Builds, tests and documents models |
| Risk, compliance, privacy, security | Second line: set standards, challenge and monitor |
| Internal audit | Third line: independent assurance |
This follows the three lines model. Internal audit must stay independent. Advising on AI governance design is acceptable; owning an AI control or approving a model is not.
A RACI matrix is good evidence of designed ownership. Approval records signed by the named owner are evidence that ownership operates.
AI policy and supporting documents
| Document | Purpose |
|---|---|
| AI policy | Principles, scope, roles, acceptable and prohibited uses, approval requirements |
| Standards | Mandatory specifics: documentation, testing, monitoring, data rules |
| Procedures | Step by step processes, such as model intake and approval |
| Acceptable use guidance | What staff may and may not do with AI tools, including public generative AI |
The auditor checks that the policy is approved at the right level, covers both built and bought AI, is reviewed when law or technology changes, and is mapped to applicable regulations.
The AI inventory
The AI inventory (or model register) lists every AI system with its owner, purpose, risk rating, data used, vendor, status and review dates. It is the audit universe for AI and the basis of compliance.
To test completeness, the auditor compares the inventory with other sources: procurement and contract records, software asset lists, cloud subscriptions, expense claims for AI tools, and network or proxy logs showing traffic to AI services. AI found in these sources but missing from the inventory is shadow AI.
Program management
A mature AI program has a pipeline from idea to retirement, with stage gates where risk, privacy, security and legal sign off. The auditor samples recent AI projects and checks that each passed the required gates, with evidence, before going live.
Program reporting should give the board KPIs (value delivered, adoption) and KRIs (incidents, models overdue for review, models outside thresholds).
Awareness, AI literacy and workforce impact
The auditor evaluates whether awareness programs align to AI policies. Training content should match current policy, reach the people who use or build AI, and be tracked for completion. Role-based training matters: developers, approvers and everyday users need different content. Good evidence is the training material compared against the policy version, plus completion records for the target population.
AI also changes jobs. The auditor may advise stakeholders on workforce impacts: which roles are automated or reshaped, how affected staff are retrained, whether new skills such as model oversight are being built, and whether changes are communicated. A risk often missed is loss of expertise: if staff stop doing a task, nobody may be able to check the AI's output.
A worked example
An insurer's AI policy says every model must be approved by the AI committee before use. The auditor obtains the inventory of 40 models and the committee minutes for the year. Only 28 models appear in the minutes. A comparison with procurement records finds five further AI tools that are in neither list.
The auditor reports two findings: the approval control is not operating for 12 models, and the inventory is incomplete. The recommendation is to bring all models through approval and add procurement and expense checks to keep the inventory complete.
How the exam asks about this
- "Which is the BEST evidence that AI governance is operating effectively?" Approval records and minutes showing models were reviewed against policy before deployment.
- "What should the auditor do FIRST to assess AI governance?" Obtain and assess the completeness of the AI inventory.
- "Who should be accountable for the outcomes of an AI model?" The business owner of the AI system, not the data science team or internal audit.
What to remember for the exam
- Every AI system needs a named, accountable business owner.
- The AI inventory is the audit universe; test it for completeness against other sources.
- A policy is design; signed approvals and minutes show operation.
- Internal audit advises and assures but never owns AI controls.
- Awareness training must match current policy and be role-based.
An approved AI policy with no evidence that projects followed it is a design without operation. Rate the governance on what happened, not on what the document says.
Common mistakes
Trusting the inventory as complete. Shadow AI is common; always reconcile with procurement, assets and network data.
Making the data scientists accountable. They build the model; the business owner is accountable for how it is used.
Ignoring the workforce. Losing human expertise weakens the oversight that controls depend on.
Check yourself
How can an auditor test the completeness of an AI inventory?
By comparing it with procurement, contract, software asset, cloud subscription and network records to find AI that is not listed.
Why is a RACI matrix not enough to prove ownership works?
It shows designed responsibility; the auditor still needs records such as signed approvals showing owners actually performed their role.
What should an AI awareness program be aligned to?
The organization's current AI policies and procedures, with content tailored to each role that uses or builds AI.
Topic 2 of 6 in AI Governance and Risk
Test yourself on AI Governance and Risk
The full ISACA AAIA question bank has 750 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by ISACA. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.