Direct Answer
Management should name the AI systems already in production and the owner of each. For the system that would cost the most if it failed, they should say what it is allowed to do without a person, what requires a human, who can halt it tonight without waiting for a committee, and who in the line business owns the result. They should also bring the errors since the last meeting, including the ones that never became incidents.
Deeper Answer
When I work with boards on AI, I want to know where a system stops recommending and starts acting, and who has the authority to intervene once it does. Companies are usually much better at explaining their AI architecture than their decision architecture.
A chatbot that drafts a customer response creates one level of risk. An agent that sends the response, changes the customer record, and issues a refund creates another. The model may be similar. The company gave it different rights.
Pick one system already in production, in customer service, pricing, hiring, healthcare, or financial decisions, anywhere a wrong action has a cost. Management should show what it is allowed to do on its own, which decisions require approval, what triggers an escalation, and who can halt the process immediately. Those answers often sit across technology, the business, legal, risk, and a vendor. Someone still has to be able to decide last.
A human in the loop can look complete in a control document while the override is slow, or punished by a metric that rewards speed. The person remains in the process. The organization has taught that person to defer to the machine.
Average accuracy tells a board very little by itself. A system can look fine overall and fail in the cases that would actually hurt the company. Drift is the same problem later: the data moved, the product moved, and nobody had already decided what happens when performance changes. Ask who is required to notice, and what they must do.
Spend the next meeting on one deployment, followed from beginning to end.
- What AI systems are in production, and who is the named owner of each.
- For the one with the largest cost if it is wrong: what may it do alone, what needs a human, and what escalates.
- Who can halt it immediately, by name, without waiting for a committee.
- What errors have occurred since the last meeting, including the ones that did not become incidents.
- When performance changes, who is required to notice, and what they must do.
- Who in the line business owns the consequence on the P&L.
Related Reading
- Your Board Does Not Need Another AI Briefing. It Needs Stop Authority.
- AI Board Governance Scorecard. 18-question assessment across six board oversight dimensions
- AI Governance Playbook for Boards
- How should boards oversee AI, and what metrics actually tell them something useful?
- If AI agents are running workflows in the background, how does a board keep control?
If the board needs this in the room, not a memo, book Liat to speak.