Artificial intelligence is becoming a central part of how companies operate. It is used in customer service, marketing, hiring, finance, and even product design. Adoption is now extremely widespread. Around 88% of organizations now use AI in at least one business function. However, governance has not kept pace. Approximately two-thirds of board directors report limited or no AI experience, and fewer than a quarter of companies have board-approved AI policies. This gap raises an important question about how effectively companies are overseeing AI decisions.
Corporate boards are regularly asked to approve AI strategies that can affect a company’s risk, growth, and long-term survival. But many board members come from backgrounds in business, law, or finance rather than technology. These skills are valuable, but AI is different from traditional business tools. It is not fixed or predictable in the same way older software systems are. AI systems learn from data, and they can change their behavior over time. This makes them harder to evaluate, especially for people without technical experience.
One problem is that AI strategies are often explained in simple business terms. Leaders may hear that AI will “increase efficiency” or “reduce costs.” While these benefits may be real, they do not always explain the risks or limitations. For example, an AI system might produce biased outcomes if it is trained on biased data. It might also generate incorrect or misleading outputs, especially in systems that create text, images, or recommendations. Without a deeper understanding of these issues, boards may approve systems that are not fully ready for real-world use.
Another challenge is how AI is presented during decision-making. Internal teams and external vendors often highlight strong results from pilot programs or controlled testing environments. These results can look impressive, but they may not reflect performance at scale. A system that works well in a test setting may behave differently when exposed to real customers, messy data, or unexpected situations.
This gap in understanding has been noted by technology leaders in the industry. Frank Palermo of NewRocket has pointed to the growing need for stronger AI literacy at the leadership level, especially as organizations move from experimenting with AI to embedding it into core business processes. The concern is not that boards need to become technical experts, but that they need enough understanding to challenge assumptions and recognize when risks are being oversimplified.
This does not mean boards are failing in their responsibilities. Many companies are starting to respond to the challenge. Some are adding directors with technical or data science experience. Others are creating dedicated technology or risk committees to focus specifically on AI oversight. There are also efforts to improve AI literacy among board members, including training on topics such as how models learn, what data they rely on, and where they are most likely to fail.
Even so, a gap still exists. AI is developing quickly, while governance practices tend to evolve more slowly. Boards may still rely heavily on management teams for technical explanations. This can become a problem if there is not enough independent review or if key assumptions are not questioned.
Speed also plays a role. Companies often feel pressure to adopt AI quickly because competitors are doing the same. This urgency can reduce the time available for careful review and discussion. When decisions are rushed, important risks can be missed, including issues related to privacy, compliance, and fairness.
To improve oversight, experts suggest treating AI not as a one-time approval but as an ongoing responsibility. Instead of simply approving a project, boards should ask how the system will be monitored after deployment. They should also ask how errors will be detected, how bias will be measured, and how models will be updated over time. In higher-risk areas, independent audits of AI systems can provide additional assurance and help validate internal claims.
In the end, boards do not need to become AI engineers. However, they do need enough understanding to ask informed questions and challenge assumptions. Effective governance depends not just on technical knowledge, but also on curiosity, structured skepticism, and a willingness to question overly simple narratives about a complex and rapidly evolving technology.
