By Chisom Obiudo
If you are a director using AI to prepare for board meetings, there is a real chance you are doing it in the one way that quietly puts your fiduciary duty at risk.
Not because AI is bad, but because nobody has taught directors how to prompt it in line with their fiduciary duties, and because AI was never built to replace your judgement.
Imagine a director opening a 150-page board pack. To save time, she ran the technical cyber risk paper through an AI tool, which returned seven brief bullet points and a comforting risk verdict: Moderate.
She reads them to the room. No one asks what the AI omitted or whether its framing aligns with the Chief Information Officer’s guidelines. The board blindly approves the paper, unaware that they have just delegated a critical governance decision to a machine’s shortcut.
Six weeks later, a ransomware attack paralyses the company’s entire network, exploiting a critical vulnerability buried on page 84 of the report that the AI’s summary completely glossed over.
The failure was not the AI. It was the prompting. It was the absence of a discipline I call AI oversight prompting: the rigorous framework a director must use to cross-examine an AI tool when the company’s future is on the line.
Productivity Prompting Versus AI Oversight Prompting
To understand why directors are getting this wrong, we need to look at how the market is currently taught to use AI. Most prompting advice online is aimed at employees who want to work faster. It tells you to assign the AI a role, provide context, show it an example or two, and specify the format you want it to return, so the AI delivers a quick, usable answer. That is productivity prompting, and the techniques are useful for working smarter, cheaper, and faster.
However, a director’s role is distinct: to exercise informed, independent judgement on behalf of the company and its stakeholders. Governance frameworks such as the NamCode and King IV call on the governing body to lead ethically and effectively. Furthermore, the duty of care, skill, and diligence remains a core common-law duty for directors under Namibian company law. None of this is suspended because a director used a tool the night before.
When directors prompt AI as analysts do, asking it to summarise and recommend, they quietly delegate judgement to a system that bears no fiduciary duty. Unlike a board, AI cannot be cross-examined or attend a disciplinary hearing.
AI oversight prompting closes that gap. It is a governance skill, not a productivity one, and it belongs on every director’s induction programme.
What AI Is, and What It Is Not
To build this oversight skill into an induction programme, a director must first look beyond the marketing hype and maintain an objective, realistic understanding of what AI can and cannot do.
AI produces text that sounds confident even when it is wrong, a failure the industry calls ‘hallucination’. It has a knowledge cut-off, so it will not know last month’s court ruling unless you feed it the data. It carries the biases of its training data, and globally trained models routinely under-represent African contexts.
It cannot be held accountable: it cannot be subpoenaed, it cannot give sworn testimony, and it will not sit beside the chair when a regulator asks how the board arrived at its decision.
AI is not a substitute for the director’s judgement. It is merely an input. The judgement, the duty, and the consequences remain entirely with the human. Treat it as a junior researcher whose work must be checked, not as a senior advisor whose conclusions can be blindly adopted.
What is AI Oversight Prompting?
From a governance perspective, AI oversight prompting is the practice of interrogating an AI system in a way that safeguards, rather than dilutes, the director’s duty of care, skill, and diligence.
Productivity prompting optimises for speed; AI oversight prompting optimises for accountability. It asks the AI to reveal what it cannot think, what it has omitted, and what a reasonable director must verify before signing a resolution.
It rests on seven core principles, each a prompt you can adapt, save, and use this quarter:
- Tie the prompt to a duty, not a topic.
Name the duty before you open the tool, or the conversation drifts into general information.
“You are advising me, a non-executive director, on management’s proposal to deploy AI to shortlist job candidates. Do not endorse the rollout. List the five questions a reasonable director must have answered before approving it, the discrimination and bias risks that management is likely to have understated, the data protection and labour law obligations the board has, and the human accountability points that must be in place before a single applicant is screened by the system. Ask me clarifying questions before you proceed.”
- Make the AI show its working.
Fluency is not truth, and a confident tone is not evidence.
“List every source, dataset, and assumption behind your last answer. Tell me how recent each one is and what you cannot know. Flag anything you are inferring rather than citing.”
- Ask for the dissent before the recommendation.
Boards fail when they hear a single narrative, so reverse the order of inquiry.
“Argue the strongest case against the position in your last answer. Write it as a memorandum from a respected dissenting director. Do not soften it.”
- Ask who and what is missing.
A globally trained model does not naturally account for local regulatory filings, community stakeholder obligations, public enterprise dynamics, or concentration risk in a small market.
“Which stakeholders, jurisdictions, and counter-scenarios did you exclude? Test your answer against Namibian conditions: NamCode, the Companies Act, and a small, concentrated market in which directors are personally known to the regulator and the public. In addition, explain your reasoning.”
- Test the recommendation against your governance code.
A recommendation that drifts from your local governance code or common-law duties is not one a director can legally adopt.
“Map your recommendation to the relevant principles of the NamCode (or King IV if applicable). Identify any principle it would breach, strain, or require the board to formally explain.”
- Ask the AI what it is not telling you.
‘I don’t know’ is the most valuable answer an AI can give a director, so force it to admit its blind spots.
“List the things you cannot reliably know about this matter. Include knowledge cut-offs, missing data, areas where you tend to hallucinate, and questions only a human inside the organisation can answer.”
- Turn the analysis into a board action.
Insight without action is not oversight. End on an operational step the chair, the company secretary, and the minutes can use.
“Convert the analysis above into three options for the board: maintain, escalate, or refer to a committee. For each, draft the resolution wording and identify the committee with primary oversight.”
Why This Matters More on African Boards
While these seven principles create a baseline defence for any director, they become an absolute necessity when operating within the specific legal and socio-economic pressures of African boardrooms.
Resist the temptation to import AI governance frameworks wholesale from foreign playbooks. African boards routinely govern AI with less internal technical capability than those in advanced markets, raising the stakes for African directors rather than lowering them. AI oversight prompting is the most accessible governance control available right now, while formal AI policies, committees, and audit functions are still being built.
Furthermore, Namibian public enterprise boards are accountable to line ministers and parliament and are subject to rigorous scrutiny, with some having community stakeholder obligations deeply embedded in their mandates as responsible corporate citizens.
A European-trained AI model will completely ignore these dynamics unless explicitly prompted to consider them. Whichever code your board reports against, the human director owns the explanation, not the machine.
Three Recommended Actions for Your Next Board Cycle
- Adopt an AI oversight prompting protocol and reflect it in the board charter, induction, and minutes. If AI is shaping decisions, the minutes must reflect that the directors independently verified the machine’s outputs.
- Log the prompts and outputs that informed your board preparation. Stating “I asked ChatGPT” is a liability, not a defence, during a regulatory inquiry. Instead, ensure your board portals maintain a real-time audit trail of the prompts used, the outputs generated, and the subsequent human verification step.
- Run the same prompt through two different AI tools. Where the answers diverge is exactly where your independent human judgement is most needed.
To translate these individual actions into collective corporate accountability, the conversation must formally begin at the table. At your next meeting, put this baseline question to the board:
“Do we have an oversight protocol for directors and committee members who use AI to prepare for board decisions, and is it reflected in our charter, minutes, and director induction?”
If the answer is no, the next item on your governance work plan has just written itself
Chisom Obiudo is an admitted legal practitioner of the High Court of Namibia and a Chief Legal Officer at the Namibian Law Reform and Development Commission. She serves on the NCRST National Artificial Intelligence Technical Advisory Committee on law and governance. Chisom holds a master’s degree in Corporate Governance and professional certificates in Non-Executive Directorship, AI Professional Skills, AI Governance and Legislative Drafting. She can be reached at: chisomokafor11@gmail.com
