The Role of NEDs in Overseeing AI-Powered Businesses

By Adrian Lawrence FCA, founder of NED Capital · Part of the Board Governance Hub

As artificial intelligence moves from a peripheral tool to the core of how many businesses operate, boards face a question they have not had to answer before: how do you oversee something you cannot fully see inside? A board is not expected to manage the technology, but it is expected to govern the company that depends on it — and when that dependence runs on AI, the oversight task changes in character. This article sets out what that oversight actually involves for a non-executive director, and how to do it well without either abdicating the responsibility or pretending to a technical expertise the role does not require.

Why AI Oversight Is a Distinct Governance Problem

Overseeing an AI-powered business is not simply overseeing an ordinary business with some new software. Two things make it distinct. First, AI systems are often opaque — even the people who build them cannot always fully explain why a model produced a particular output, which makes the “black box” problem a governance issue rather than merely a technical one: a board cannot hold to account what it cannot understand. Second, AI can fail in ways that are unfamiliar and fast-moving — a model can be quietly biased, can degrade as the world changes around it, or can produce confident but wrong outputs at scale before anyone notices. These are not the risks traditional board oversight was designed around. The consequence is that a board overseeing an AI-powered business needs to adapt how it governs: not by becoming technical, but by insisting on the transparency, accountability and control frameworks that make an opaque, fast-changing technology governable at all. Recognising that AI oversight is a genuinely new kind of problem — not a familiar one with a new label — is the starting point for doing it well.

The Questions a Board Should Be Asking

The most practical way a non-executive director exercises AI oversight is through the questions they ask. A board does not need to audit a model to govern it well; it needs to know that the right questions are being asked and answered credibly. Does the board actually know where and how AI is used across the business, including in ways that have crept in without formal decision? Who is accountable — is there a clear owner for AI risk, rather than a diffuse sense that “the tech team has it”? How are the systems tested before and after deployment, and how would the company know if a model started behaving badly? What data are the models trained and run on, and is its use lawful and sound? And critically, what happens when the AI is wrong — is there human oversight of consequential decisions, a way to catch errors, and a plan for when something goes wrong at scale? A NED who consistently presses these questions, and refuses to accept reassuring vagueness in reply, is doing the core of the oversight job. The questions matter more than the answers being perfect: their absence is the warning sign.

The Risks That Need Board Attention

Several categories of AI risk deserve a place on the board’s agenda, and each is serious enough to warrant focused attention rather than a passing mention. Bias and fairness — the danger that a model reproduces or amplifies discrimination present in its data — is both an ethical and, increasingly, a legal and reputational risk; it is examined in the NED’s role in monitoring AI bias and ethical risk. The broader ethical dilemmas of adopting AI — where the commercially attractive option and the responsible one diverge — are explored in how NEDs can manage ethical dilemmas in AI adoption. Data governance, cybersecurity and operational resilience take on added weight when a business depends on AI systems and the data feeding them. And there is the longer-horizon question of how far a board should be preparing for more capable AI still to come, which is taken up in whether boards should prepare now for AGI risks. The board’s task is not to solve each of these itself, but to ensure each is owned, monitored and reported on properly — and to know which of them, for its particular business, is most likely to cause serious harm.

Oversight Without Micromanaging

A recurring tension in AI oversight is the same one that runs through all non-executive work, only sharper: how to hold management genuinely to account without straying into managing the technology yourself. The line matters. A non-executive director who tries to direct technical decisions oversteps the role and is usually out of their depth; one who defers entirely to the executive team on the grounds that AI is “too technical” abdicates it. The right posture sits between: the board sets the expectation that AI is developed and deployed responsibly, insists on the frameworks and reporting that let it verify this, and challenges hard when the answers are thin — but leaves the building and running of the systems to those whose job it is. This is oversight, not operation, applied to a domain where the temptation to either over-reach or opt out is unusually strong. Getting that balance right is much of what distinguishes effective AI governance from either interference or neglect, and the general principle is developed in how NEDs create accountability without micromanaging.

Does the Board Have the Right Skills?

All of this raises a question about the board’s own composition. A board overseeing an AI-powered business does not need to be full of data scientists, but it does need enough genuine AI literacy to ask good questions, interpret the answers, and know when it is being reassured rather than informed. A board with no one who understands how AI systems fail is poorly placed to govern a company that depends on them — it cannot challenge what it cannot follow. For many boards, this means either developing the existing directors’ understanding or, increasingly, appointing a non-executive director who brings real fluency in technology and its governance. That does not mean a narrow technologist; it means someone who combines credible AI understanding with the broader judgement of a good board member, able to translate between the technical and the strategic. Building that capability into the board is itself a governance decision worth making deliberately rather than by default. At NED Capital we help boards find non-executive directors with genuine technology and AI-governance fluency, alongside the judgement the role demands. Every search is led personally by Adrian Lawrence FCA, a Fellow of the ICAEW and former listed-company finance director.

NED Capital | Sister practice of FD Capital | ICAEW practising certificate held by Adrian Lawrence FCA.