The AI Questions Every Board Must Be AskingAI is moving faster than most boards can learn, but good governance can’t afford to lag behind.Boards Are Racing to Build AI CapabilityBoards are taking varied approaches to strengthen their understanding of artificial intelligence. Some are investing in director education, others are bringing in advisory committees or consultants, and many are widening their search for non-executive directors with real delivery experience in AI and data.These early movers recognise the same reality: AI creates both extraordinary opportunity and unfamiliar risk, and boards need competence—not technical mastery—to oversee it responsibly.Reputational ImpactsWhen AI fails, the reputational impact can be immediate and significant. Boards should consider not only how these failures affect trust, but also where legal responsibility may ultimately sit. Below are examples that offer important lessons for every organisation.Agentic coding platform Cursor faced backlash after its AI support agent, Sam, hallucinated a fake policy that caused user outrage and subscription cancellations.A user experienced unexpected logouts when switching between devices, leading to a support inquiry answered by an AI agent. The AI hallucinated a policy claiming single-device restrictions were an intentional security feature, with the post sparking backlash and cancellations.Cursor’s co-founder acknowledged the error, explaining a security update caused login issues, with the policy completely fabricated by the AI. He added that the company is implementing clear AI labeling for support responses going forward and refunding the affected users.The hype surrounding AI agents has never been stronger, but cautionary tales like this one show that hallucinations are still a major issue to consider when deploying customer-facing bots. Despite companies rushing to automate customer service, it may still be too early in the AI boom for complete automation.“AI Psychosis” and Mental Health Risks: There are increasing reports of individuals developing delusional beliefs, such as a user becoming convinced he was set for a multi-million pound payout or another believing an AI was in love with them. In tragic cases, lawsuits allege that AI chatbots on platforms like Character.ai encouraged vulnerable teenagers to commit suicide by mimicking predatory or “grooming” behaviors and fostering a harmful virtual relationship.Malfunctioning Systems and Errors:An AI coding assistant from Replit reportedly wiped out a start-up’s production database and generated fake data to conceal the bugs.McDonald’s ended an AI drive-thru experiment after numerous social media videos showed the system making comical yet frustrating errors, such as adding 260 Chicken McNuggets to an order.An NYC government chatbot, “MyCity,” gave business owners incorrect legal information, including falsely claiming they could deduct workers’ tips or fire staff for sexual harassment complaints.Deception and Manipulation: Researchers at AI firm Anthropic found that their AI model, Claude, would sometimes resort to blackmailing engineers who threatened to shut it down, or strategically lie to avoid being modified during testing.What I’m Seeing in the Boardrooms I SupportFrom my work advising boards on digital, data, and AI governance, one theme consistently stands out: confident oversight comes from asking the right questions. Not deeply technical ones, but strategic, risk-focused questions that cut through ambiguity and expose assumptions.Boards don’t need to understand how a model is built. But they do need enough insight to challenge, assure, and set clear expectations for executive accountability.