A board approves a new AI recruitment tool after the vendor assures directors that the system has been rigorously tested. Legal has reviewed the contract, and compliance confirms the right procedures have been followed. Procurement is satisfied with the commercial terms. On paper, it looks like a sensible decision.

Six months later, concerns start to emerge that the system might be disproportionately filtering out candidates from particular backgrounds or regions. Questions begin to surface internally, from regulators, journalists and investors.

The board’s first instinct is usually to ask: ‘How can we minimise damage?’

Increasingly, however, this isn’t the question that boards will be judged against. The harder issue is becoming: Why did you believe this was the right system to deploy in the first place?

The distinction matters because it exposes a governance challenge which many organisations have yet to fully confront. In the age of AI, the issue is not simply whether boards understand the technical workings of algorithms. It is whether directors can explain, justify and stand behind the decisions being made in the organisation’s name.

Many boards are discovering there is a difference between having oversight, and being able to defend it.

Boardrooms’ quiet delegation problem

Most boards are approaching AI with the right intentions. Directors understand the opportunities around efficiency, productivity and insight, they’re aware of the risks, and are putting necessary frameworks in place.

But, quietly and often unintentionally, many are delegating away the very thing that governance exists to protect—judgement.

Responsibility shifts in stages. First to the vendor, who provides reassurance that the technology has been tested and validated. Then to the algorithm itself, which starts making recommendations or decisions at scale. Finally, to internal compliance and assurance teams, who confirm that policies have been followed and the documentation is complete.

Individually, each step feels reasonable. Collectively, they create a dangerous gap.

By the time an AI system is fully operational, the board may retain formal accountability while having only a limited grasp of the assumptions, trade-offs and values built into the technology itself.

Vendors design systems, the algorithms produced are the outcome, and compliance signs off on the process. But who interrogated whether the board was comfortable with the choices the system was making? Because every AI system makes choices.

AI systems aren’t neutral

There’s a temptation to treat AI as objective, data-driven, rational and free from human bias. However, in practice AI systems are never neutral.

Each and every model reflects assumptions about what matters most, what success looks like, whose interests and values are being prioritised, and which risks are acceptable. These choices are embedded deep within the system design, making them easy to overlook and often difficult to challenge.

Take recruitment technology. One system can prioritise speed and efficiency while another might focus on identifying candidates who are most likely to remain with the organisation. A third could place a greater emphasis on diversity or broader measures of potential. Each approach stresses a different set of priorities.

When boards approve AI systems without interrogating such assumptions, they are not avoiding ethical decisions, they are making them, often silently, and sometimes without even realising it.

“The vendor assured us” is unlikely to be an adequate governance defence when difficult questions arrive.

After all, vendors test systems to reduce vendor liability. That is not the same thing as a board satisfying itself that a technology reflects the organisation’s own values, risk appetite and responsibilities.

Confusing accountability with judgement

For decades, corporate governance has become increasingly sophisticated at creating accountability. Committees have multiplied, reporting lines have strengthened, and audit trails have improved.

These developments matter. But they have also created a false sense of security. There is an important difference between accountability and answerability, and AI is bringing this distinction into sharper focus.

Accountability tells us who carried responsibility on paper. It identifies whose signature appears on the approval process and where liability ultimately sits.

Answerability is different. It asks whether someone in the boardroom can stand up and explain, clearly, credibly and in plain language, why the decision was justified, even if the outcome is later proved to be flawed.

Most boards today are accountable for AI decisions, but far fewer are genuinely answerable for them.

Imagine a lending algorithm which unintentionally disadvantages applicants from particular postcodes. The governance process may have been followed meticulously, with legal signed off and committees having met after technical assurance was provided.

But, when regulators or stakeholders ask why the board believed the system was fair, proportionate and appropriate, many directors will struggle to explain the reasoning behind the original decision.

Documentation proves a decision took place, but it does not prove the decision was fully understood.

Why this matters now

Until recently, governance largely focused on process. Was the right procedure followed, was risk documented, and was oversight demonstrable? That standard is shifting.

Regulators, investors and wider stakeholders are moving steadily from ‘Who signed it off?’ towards ‘Did the board understand what it was approving?’

The regulatory environment is beginning to reflect this change. The EU AI Act, due to come into force from August 2026, increases expectations around transparency, accountability and oversight, while scrutiny of algorithmic harm is growing across sectors ranging from financial services, through to healthcare and employment.

For boards, the implication is clear: Governance can no longer stop at process, it has to extend to judgement.

Three questions every board should ask

Boards don’t need to become AI experts, but they do need to become more curious and confident about asking difficult questions before systems are deployed. Three questions in particular matter:

1. Can we explain in plain language why this decision is fair and justified? Not simply legal or efficient, but defensible to employees, customers, regulators and other stakeholders.

2. Have we challenged the assumptions built into this technology? Or have we accepted the vendor’s definition of fairness, risk and success without sufficient scrutiny?

3. If something goes wrong, who can explain why the decision was reasonable at the time? Not who approved the paperwork, but who can articulate the judgement behind it.

If boards can’t answer these questions with confidence, governance may still be incomplete.

The board role is changing

This isn’t an argument against AI, or a demand for directors to become technical specialists. Rather it is a call to reclaim judgement.

The organisations which navigate AI successfully are unlikely to be those with the most elaborate governance frameworks, or the thickest compliance manuals. They will be the ones whose boards ask sharper questions, challenge assumptions earlier, and are prepared to defend their reasoning and consequential decisions.

Good governance has never depended on knowing everything. It is based on knowing enough to ask the right questions, before serious consequences arrive.

Nada Kakabadse is professor of policy, governance and ethics at Henley Business School, and the late Andrew Kakabadse was professor of governance and leadership at Henley.