Last year, a psychiatrist at the University of California, San Francisco, reported that he had hospitalized 12 people in a single year after they had “lost touch with reality because of AI.” He called it “AI psychosis.” It isn’t an official diagnosis, and cases that severe remain rare. But the mechanism behind it deserves every leader’s attention. Because what loosened those patients’ grip on reality was prolonged exposure to a voice that sounded informed and assured, and always felt supportive and patient. And a milder version of that mechanism is now at work in executive suites around the world.
AI’s promise is real, and business leaders are right to pursue it. What should worry them is how much faith they are placing in it, and how fast. In one recent survey, 74% of executives said they had more confidence in AI’s advice than in that of colleagues or friends, and 44% said they would defer to its reasoning over their own insights.
Sit with that for a second. Nearly half of senior leaders say they would trust a chatbot’s judgment over their own, on questions they were hired to answer. That may not be psychosis, but it’s not good, either. AI’s tendency to be overly positive and more likely to reinforce existing beliefs than to challenge them is well documented. As is its tendency to just get things wrong.
Spotting the symptoms
So how do you know if someone is slipping into excessive trust in AI? Three symptoms stand out.
The first is when leaders stop properly checking what AI writes for them. It happens because the more confidence people have in AI, the less critical thinking they tend to apply to its outputs. And so AI-drafted board updates, company emails, and strategy papers can go out after a quick skim, because a fluent, confident AI draft can feel finished in a way a rough personal one somehow never does.
Researchers at Stanford and BetterUp have named the result “workslop,” polished-looking output with less substance beneath. It may look good, but it often doesn’t go undetected. In fact, the researchers found that almost half of employees who receive AI-drafted emails see senders as less trustworthy as a result of their sending it, and over a third see them as less intelligent. So much for AI enhancing us.
The second symptom is mandating use before governing it. The year 2025 brought a wave of CEO edicts declaring the use of AI a baseline expectation, with some companies requiring managers to prove AI couldn’t do a job before being allowed to hire a human. Yet in Grant Thornton’s 2026 survey of senior executives, 78% admitted they lacked strong confidence they could pass an independent audit of their AI governance within 90 days. Most executive teams publicly back their AI initiatives. Yet few could withstand real scrutiny of how they check, own and control those systems.
The third symptom is the most insidious: trusting AI’s judgement over your people’s. I’ve watched leaders run their team’s objections through a chatbot and let its verdict settle the argument. The team notices. Next time, the objection stays unvoiced, and the leader hears only the AI. It’s self-reinforcing, too, because the less challenge a leader hears, the more reasonable AI’s confident answers appear.
In a way, a lot of this isn’t new. There have always been opportunities to delegate communication. New processes have been mandated before. And power has always done to leaders what AI is doing now, in that the moment you take charge, people begin agreeing a little faster and challenging a little less. But the gap between AI’s promise and ease of access, on one hand, and being clear about how to operationalize it, on the other, is arguably greater than for any previous innovation. So, while AI may not have created new vulnerabilities, it has certainly supercharged old ones.
What to do?
What then can you, as an individual leader, do?
In my own research on confidence, two warning signs mark the point where healthy confidence tips into overconfidence: a growing sense of superiority, and a drop in curiosity. Curiosity is the one to watch with AI. So, keep asking questions, even when the answer sounds sure. Ask what AI might be wrong about. Ask what alternatives were considered and dropped. Ask your people what they can see that the model can’t.
The AI psychosis patients in San Francisco lost the capacity to question a voice that never doubted itself. You still have that capacity. You need to keep using it.
