AI expertise isn’t the same as AI oversight. Boards need both

From the second quarter of 2023 to the second quarter of 2025, the proportion of S&P 500 firms reporting AI as a risk jumped from 12% to 83%. Yet, despite an increase in disclosures, only a small fraction of board members possess AI expertise. In the research conducted, only 23% of executives declared their board to be highly knowledgeable in AI, and a quarter described their board's AI fluency as low or nonexistent.

As AI becomes more integrated into business operations, this discrepancy between understanding and oversight becomes substantial, posing a significant challenge. This issue is inescapable, and boards must address it to fulfill their responsibilities. Expertise is necessary, but not sufficient Boards require both AI expertise and oversight, but simply seeking out AI experts is insufficient to bridge the gap.

There are two reasons for this. Firstly, AI is not isolated; it permeates various aspects of business, including hiring, pricing, customer service, credit, and operations. No single expert can own the AI agenda, and every director will inevitably vote on matters influenced by AI, whether they realize it or not. Therefore, oversight must be distributed across the board rather than concentrated in a few expert positions.

Secondly, expertise and competent oversight are not synonymous. While expertise is the knowledge about AI, oversight involves the actions taken based on that knowledge. Knowledge alone does not guarantee effective oversight. Proficiency in AI is necessary for competent oversight, but it is not sufficient. For instance, understanding how a large language model operates does not necessarily require knowledge of who within the company is responsible for it or the permissions granted to AI agents.

Consequently, boards face two critical questions about AI: whether the board possesses sufficient knowledge, and whether each director is competent in AI oversight. The first question addresses knowledge, while the second focuses on the board's ability to oversee effectively. The board's responsibilities in AI oversight Boards must now determine what constitutes effective AI oversight.

Drawing from ongoing research into how directors oversee AI, the following foundational areas emerge, each representing a critical aspect that requires board attention to prevent potential harm to the company. Governance Oversight: Identifying the individual accountable for each AI matter and ensuring they are held accountable. Data Inquiry: Understanding the data sources used by the company's AI, where the data is stored, and the legal basis for its usage.

This inquiry must be substantiated by evidence rather than vendor assurances. Explanation & Disclosure: Ensuring the company can explain AI decisions to affected individuals and that its public statements about AI align with its actual practices. Fairness & Affected Parties: Recognizing groups that could be disadvantaged by AI systems, the frequency of testing, and the effectiveness of addressing identified issues.

Risk & Irreversibility: Assessing the capabilities of the company's AI that cannot be reversed and establishing a mechanism for board notification of failures. Human Judgment & Workforce: Confirming that human intervention can override AI systems in practice and that employees whose work is affected by AI are involved in the oversight process.

Autonomy & Agent Authority: Understanding the actions, spending, and decision-making capabilities of AI agents, ensuring they operate within predefined limits and have mechanisms for control. Building competent oversight of these areas involves three key principles. Measure performance, not just knowledge The most reliable evidence of a director's ability to perform AI oversight tasks is their demonstrated performance in those tasks.

Therefore, it is essential not only to assess directors' theoretical understanding of AI oversight but also to inquire about their recent practical experience. How often have they performed basic AI oversight tasks, and is it integrated into their regular oversight processes? The distinction between awareness and execution, or being good at performing a task, is crucial.

Despite two-thirds of surveyed directors expressing the need for increased AI discussion on the board, fewer than half acknowledged that their board actively oversees AI. For example, while most directors would agree that boards should be aware of the AI systems in use, fewer have directly requested and reviewed these systems or regularly assess their use.

The first step is for each director to identify their most recent AI oversight action, including the date and context of this action. Assessing the lowest competency level is essential, not just the average. This approach aligns with financial control standards, where a single material weakness can render an entire system ineffective, regardless of the performance of other controls.

Boards should adopt a similar approach for AI oversight. The floor sets the minimum standard for all directors. Assess the floor, not the average Each director's competence is determined by their weakest area of AI oversight, not their average performance. Boards must acknowledge this principle in financial controls, where a material weakness in one area would render the entire control ineffective, despite strengths in other areas.

AI oversight requires the same rigorous approach. The floor is not a demand for directors to become experts in all aspects of AI. Rather, it sets a minimum standard that every director must meet in every area. Boards already apply this principle in finance, requiring every member of the audit committee to possess financial literacy.

Competent AI oversight requires a solid foundation: a floor cleared by every director. First step: ask each director which AI oversight question they would be least capable of answering.