See each use
Collect AI uses in one shared inventory and name an owner for each of them.
Approaches / Governance
Turning “Who governs AI?” into concrete authority.
Ask an organization who governs AI and you may hear “the IT team”, “we have a committee” or “our policy is published”. Make the question concrete and the model appears. Who approves a new use case, who can stop a live system, and who is accountable for the outcome?
Governance works not because a policy is published, but because it is clear up front who proposes, approves and can stop each decision. The approach therefore begins not with a policy, but with decision rights and risk limits.
The governance movement
Collect AI uses in one shared inventory and name an owner for each of them.
Weigh expected business value, risks and the people affected together, and place each use in a risk class.
Define who proposes, who approves and who can stop each decision, and within which risk limits they act.
Carry evidence that the limits hold in practice to the board through regular reporting.
The four-layer model
How the four layers map onto existing structures depends on the organization’s scale and risk profile.
The board approves the risk limits. Evidence that those limits hold in practice reaches the board through regular reporting.
Chain of decisions
↓ Decision rights flow down↑ Evidence flows up
Common spine
The board oversees how AI use serves business strategy and how significant risks are managed. It approves risk appetite and tracks the results of implementation through regular reporting.
An AI Governance Council, representing senior management and the relevant business units, prioritizes the portfolio of uses, oversees risk classification and decides high-risk uses within delegated authority. Decisions beyond its authority go to senior management.
A business owner is accountable for each use across design, testing, go-live and monitoring. Technology and data teams support delivery; a central AI office, where one exists, provides shared tools, methods and expertise.
Risk and compliance teams set control requirements and monitor implementation; legal and information security contribute to decisions in their areas. In the Three Lines Model, risk and compliance sit in the second line and internal audit provides independent assurance as the third.
Decision rights
Weigh each use’s expected business value, its risks and who it affects together. Organizational values and legal, professional and ethical principles are among the criteria. For high-impact decisions, define in advance when an authorized person steps in and who may change or stop the decision.
| Decision | Proposes | Approves | Can stop |
|---|---|---|---|
| New low-risk use | Business owner | Business unit head | Business owner or authorized risk officer |
| High-risk use | Business owner | AI Governance Council, within delegated authority | AI Governance Council or authorized risk and compliance officer |
| Enterprise-wide AI policy | AI Governance Council | Senior management. The board is informed. | Senior management may suspend its entry into force |
| Supplier and contract approval | Business owner and technology team | Authorized procurement manager, with legal and information security input | Authorized legal or information security officer, on a control breach |
| Incident response on a live system | Incident response team recommends restart | Business owner approves restart after controls | Business owner or authorized incident response lead |
| Risk appetite and limits | AI Governance Council | Board of directors | Board of directors may halt a change to the limits |
This table shows an illustrative allocation of authority. Whether a model is suitable for a specific use is decided on the low-risk or high-risk use row, according to that use’s risk class. Anyone who sees a problem in production can report it; authority to stop and restart belongs to roles defined in advance.
The first 90 days
Put uses into the inventory, name an owner for each, and draft the core principles.
Define risk classes, decision rights and the working rules of the AI Governance Council.
Review the first high-risk uses and present the first oversight report to the board.
The test question
“Who decides to go live, and under what conditions who can stop it?”
Ask it when the first high-risk use is ready to go live. If the teams give the same answer, the model has started to take hold in daily operations.
What the frameworks say
The KPMG International and INSEAD framework for boards organizes AI oversight around five principles: strategic oversight, active technology and security oversight, workforce transformation and human accountability, trustworthy AI, and the board’s role and operation.
In the Three Lines Model the governing body takes accountability, management runs the first- and second-line roles, and internal audit provides independent assurance. The four-layer model adapts that separation to AI.
NIST’s AI risk management framework has four functions: govern, map, measure and manage. Govern is a cross-cutting function that runs through the other three.
Regulation (EU) 2024/1689 classifies AI systems by risk level and adds obligations for high-risk systems. That is why tying risk classification to decision rights matters.
Where to start
Assessment
See how far data governance, information security, model risk, personal data protection and supplier management already cover AI use, and set up a shared inventory and risk classification.
Suitable when: AI use is spreading but ownership is scattered.
Decision rights
Adapt the decision-rights table to the organization and write the working rules of the AI Governance Council.
Suitable when: the first high-risk uses are approaching go-live.
Board oversight
Design the board oversight dashboard and the first oversight report, and settle what evidence the board sees and how often.
Suitable when: the board is responsible for AI oversight but does not yet receive regular evidence.
Shared benefit
Success is measured not by publishing a policy but by whether teams give the same answer on go-live and stop decisions, and whether the board receives regular evidence.
If you would like to talk it through, leave a note. LinkedIn ↗ · Reflections →