Reflections on "Power Steering, Not a Brake: How Boards Should Actually Govern AI"
Artificial intelligence is rapidly becoming a standing item on board agendas. Yet many directors remain uncertain about what their role should actually be.
Should boards be approving AI tools? Challenging algorithms? Hiring AI specialists? Creating dedicated AI committees?
A recent paper, Power Steering, Not a Brake: How Boards Should Actually Govern AI, offers a refreshingly practical answer. Rather than treating AI as a separate governance discipline, the authors argue that AI should be governed through the same responsibilities boards have always held: purpose, strategy, risk, capability, and leadership.
That may sound obvious. In practice, it is one of the most important insights in the entire paper.
The real risk is not AI. It is governance failure.
Many organisations are currently making one of two mistakes.
Some boards remain under-engaged. They recognise AI is important but lack the confidence or understanding to challenge management effectively. As a result, investment decisions receive limited scrutiny and opportunities are missed.
Others move to the opposite extreme. Enthusiastic directors push management to deploy AI rapidly without fully understanding the underlying data quality, operational readiness, regulatory exposure, or control environment.
The paper describes these as the "Clueless Board" and the "FOMO Board."
Most organisations, arguably, are suffering from both conditions simultaneously. They are moving too slowly to capture value and too quickly to control risk.
That observation should resonate with many directors. The challenge is rarely a lack of awareness. It is a lack of governance architecture.
The most valuable question: Where does human judgement end?
One of the most thought-provoking sections of the paper focuses on purpose.
AI systems are exceptional optimisers. They can process vast quantities of information and make recommendations at a scale no human team could match.
However, optimisation is not judgement.
An AI system can optimise for efficiency, profitability, productivity, or customer acquisition. It cannot determine whether those outcomes are aligned with an organisation's purpose, values, or responsibilities.
This distinction matters.
The decision about what should be automated, what should remain human, and where accountability ultimately resides remains a board-level question.
As AI capabilities become more sophisticated, the importance of human judgment does not diminish. It becomes more valuable.
Why the STAR framework deserves attention
The paper introduces a governance tool called STAR:
- Shareholder value thesis
- Threat parity
- Ability
- Risk budget
What makes STAR useful is its simplicity.
Boards do not need to understand neural networks, model architecture, or prompt engineering. They do need to understand whether an initiative creates value, whether controls are keeping pace with emerging threats, whether the organisation can realistically execute, and whether accountability is clearly defined.
The framework provides a structure for those conversations without requiring directors to become technologists.
In many respects, STAR performs the same role for AI oversight that audit committees perform for financial reporting: it creates a repeatable discipline for asking the right questions.
A particularly important insight: governance as an enabler
Perhaps the strongest message in the paper is contained in its title.
Many organisations still treat governance as something that slows progress.
The authors argue the opposite.
Strong governance should function as power steering, not a brake.
A well-governed organisation can deploy innovation more confidently because responsibilities are clear, controls are tested, escalation routes exist, and decision-makers understand their boundaries.
Poor governance creates hesitation because nobody knows where accountability sits.
For boards, this is a useful reminder that effective oversight should accelerate responsible decision-making rather than obstruct it.
The emerging board capability gap
Another important theme is the growing information imbalance between boards and management.
Historically, directors could often draw upon decades of industry experience to challenge executives directly.
AI changes that dynamic.
Management teams typically possess far greater technical understanding than their boards. This creates a dependency that many governance models were never designed to accommodate.
The answer is not for every director to become an AI expert.
It is for boards to invest continuously in learning, seek independent perspectives, and develop enough understanding to challenge assumptions rather than simply accept them.
The board's role remains oversight, but oversight becomes impossible without a working level of literacy.
The question every board should ask
The paper closes with a deceptively simple challenge:
Can your board answer its key AI governance questions with evidence?
Not optimism.
Not confidence.
Not policy documents.
Evidence.
Can management demonstrate value creation? Can it demonstrate control effectiveness? Can it identify accountable owners? Can it explain where AI risk is acceptable and where it is not?
Boards that can answer those questions are likely to navigate the AI era successfully.
Those that cannot may discover that the greatest AI risk was never the technology itself, but the absence of effective governance around it.
NEDness view
The most important lesson from this paper is that AI governance is not fundamentally about artificial intelligence.
It is about how boards exercise judgement when technology changes faster than organisational structures.
The directors who create the greatest value will not necessarily be those with the deepest technical expertise.
They will be those who remain curious, ask better questions, insist on evidence, and ensure that innovation advances within a framework of accountability.
That has always been the essence of good governance. AI simply raises the stakes.