Boardroom Blind Spots: Nick Curum Highlights AI Accountability Gaps

In a recent LinkedIn post, Nick Curum highlights a critical and often overlooked issue in corporate governance: the widespread, unapproved use of AI tools by senior leadership and the subsequent lack of oversight.

Curum’s analysis, based on a survey of over 1,000 employees, reveals a startling statistic: 93% of senior managers and executives are utilizing AI tools without explicit organizational approval. This behavior, he emphasizes, is not confined to junior staff but is prevalent among those in decision-making roles.

“It is not an edge case. It is the default behaviour at the top of most organisations right now.”

This widespread adoption of unvetted AI tools by leadership prompts Curum to question the current focus of board discussions. He asserts that boards are frequently asking the wrong questions, focusing on whether an AI strategy exists rather than on the immediate, tangible impact of AI on decision-making.

The Wrong Question vs. The Right Question

Curum argues that the fundamental inquiry for boards should shift from a high-level strategy discussion to a granular examination of AI’s current operational influence. The critical distinction, according to Curum, lies between asking about an AI strategy and demanding to know “what AI is currently deciding, and whether anyone in the boardroom has visibility of it.” These questions, he notes, have vastly different implications for risk and accountability.

Key Questions for Boardroom Accountability

To address this governance gap, Nick Curum proposes four crucial questions that should be central to board and risk committee meetings:

1. What is AI allowed to decide today?
2. Where is it influencing decisions right now?
3. Who reviews AI decisions before action?
4. What controls AI outputs?

Curum suggests that silence, or a reference to unread documentation, in response to these questions signals a significant accountability gap, not a technological one.

“If any of those produce silence, or a reference to a document nobody has read, you have located the gap. Not in the technology. In the accountability structure around it.”

Deconstructing AI Oversight: Curum’s Four-Layer Framework

The post introduces a four-layer framework designed to map and manage AI oversight. Each layer represents a critical control point with identifiable failure signals and executive checks:

Layer 1: System

This layer focuses on how AI outputs are generated and validated. The failure signal here is the use of outputs without validation, such as actioning AI recommendations directly without any review or record of approval.

Layer 2: Process

This layer ensures AI decisions are visible before action is taken. A failure occurs when there’s no formal review process, leading to informal, inconsistent practices depending on the user.

Layer 3: Operating Model

Here, the control is over where AI influences decisions. AI being used outside approved workflows without updating risk registers or informing the board is a key failure signal.

Layer 4: Governance

This foundational layer defines what AI is permitted to decide. A failure occurs when a broad AI strategy is approved, but specific decision rights, especially those without human intervention, remain undefined.

“The organisations that handle this well are not more advanced technically. They are more precise about accountability.”

Curum concludes that organizations excelling in AI governance are distinguished by their precision in accountability, understanding which layer they operate within and who owns each component. This clarity, he argues, separates genuine oversight from mere compliance documentation. He encourages leaders to proactively assess their own organizations’ positions regarding these critical AI accountability questions.