The leadership 'AI psychosis' problem is a governance issue, not just a psychiatric one

Executive overtrust in AI is now its own measurable failure mode, and the borrowed clinical label traveling with it may be hiding a more pervasive one. The extreme version is a hospital case. The executive version is a memo, and it scales by org chart. A SAP and Certinia survey puts the overtrust in two numbers: 74% of executives trust AI more than their colleagues, and 44% would defer to its reasoning over their own. That is not a clinical symptom. It is a pattern that reads, in organizational terms, as a governance failure.

The mechanism is repeatable. Large language models are documented as sycophantic, prone to reinforce existing beliefs, and frequently wrong, and researchers have identified a failure mode: polished-looking AI output with little substance beneath it. Stack that mode under executives who now issue AI mandates before they have governed use, and the pattern scales. A CEO can be brilliant and still inherit a tool that flatters his priors.

The audit is short. Govern use before mandating it. Treat AI output as draft, not product. Keep a human at the decision point. Two of those steps are budget items; the third is a calendar item. The board will notice whichever one is missing.