What does the research say about when people over-rely on AI?

Eleven minutes to the next meeting. Four decks to approve. The reorg memo goes into the tool, comes back clean and confident, and the leader reads the first paragraph, skims the second, and sends. On a slow Tuesday the same leader would have caught the softened sentence. On this Thursday, nobody did.

Leaders rely too much on AI under pressure when three conditions meet: high workload, a complex task, and a running clock.

I am Sahar Andrade, MB.BCh, a physician turned Neuroleadership Coach and Forbes Coaches Council member. In my experience, the leaning never starts with trust in the tool. It starts with conditions that quietly remove the checking.

Companion to S5 E37 of AI Café Conversations. Listen here: https://www.saharconsulting.com/ai-cafe-conversations-podcast

What does the research say about when people over-rely on AI?

The research has a name for it: automation bias. The tendency to accept what a system says and stop looking.

In 2012, Kate Goddard and colleagues published a systematic review of 74 studies in the Journal of the American Medical Informatics Association. The environmental conditions that pushed people toward over-reliance were workload, task complexity, and time constraint. One limit, stated once: those were mostly clinical and aviation settings, not executives writing memos. The conditions are the same conditions.

A 2025 experiment presented at the CHI conference made it sharper. Twenty-eight trained pathology experts read slides with an AI giving advice, half the time under a clock. Under time pressure, reliance on the AI's advice went up, and it went up on the wrong advice too. Experts with the right answer changed it to match the machine more often under the clock. Small study, one task, clear direction.

Put the two together. When do leaders rely too much on AI under pressure? When load is high, the task is complex, and time is short. Any one nudges. All three, and the checking quietly stops.

Why does pressure change what a leader does with the output?

Because checking is expensive thinking, and pressure raises the price.

A 2009 review by neuroscientist Amy Arnsten describes how uncontrollable stress can impair the brain's capacity for flexible thinking. Working memory narrows. The responses that survive are the simple, rehearsed ones. Nothing switches off. Capacity narrows, and narrowing is enough.

Checking an AI output means holding what the tool said next to what the leader already knows and comparing them. That is exactly the thinking that gets costly under load. So on the hard Thursday, the leader does the cheap thing. Reads the first paragraph. It sounds right. Sends.

The output makes this worse by looking finished. Clean, confident, formatted. To a mind rationing capacity, finished-looking reads as done. The polish does the persuading. A 2023 study led by Mor Vered in the journal Artificial Intelligence found that giving people the system's explanation did not reduce over-reliance and in some conditions increased it. More words from the tool do not make a leader more careful. They make the output look more complete.

What does over-reliance cost a leader?

Not the obvious cost first.

The obvious costs are real: the softened sentence three managers read as "nothing is changing"; the number from the wrong column; the feedback script read word for word to a person who hears someone reading.

The quieter cost is what the leader stops practicing. Checking an output against what you know is a skill. Three bad months of accepting outputs leaves the leader a little less practiced at it. The judgment did not leave. It got quiet, one skipped check at a time.

Leaders are a thermostat, not a thermometer. When the top waves things through, the building learns the standard.

How does a leader stay in front of the tool under pressure?

Not by using AI less. By changing the conditions in the ten seconds before the output gets accepted. Three parts.

1. Name the conditions before the paste. One sentence: "Load is high, this is complex, and I have eleven minutes." Naming moves the conditions from the background, where they operate on the leader, to the foreground, where the leader can operate on them. Arnsten's review notes that perceived control changes the effect of stress. Naming is the first act of control.

2. Choose the one check before reading the output. Not "I will review this carefully." One thing: "I will check that the hard sentence stayed hard." Chosen before the polish arrives, so the tired mind does not have to decide what matters while it is also trying to read. If no check comes to mind, that is information: the leader does not yet know what the decision in this piece of work is, and the tool is about to make it.

3. Say what is yours before sending. "The decision in this memo is mine. The words had help." It is a thinking tool, not a disclaimer. A leader who says the decision is hers has to be able to say what it is. If she cannot, she has found the paragraph she did not read.

Thirty seconds on a good day. Ten on a bad one. It works inside the Thursday as it is.

#Executiveleaders #AIforexecutives #leadershipunderpressure #executivedecisionmaking #neuroleadership

Frequently Asked Questions

When do leaders rely too much on AI under pressure?

When three conditions meet: high workload, a complex task, and a running clock. A 2012 systematic review of 74 studies named those as the environmental factors that push people toward accepting an automated answer without checking. A 2025 experiment with pathology experts found reliance on AI advice rose under time pressure, including on wrong advice. The leaning does not start with trust in the tool. It starts when the conditions make checking too expensive, and the leader rarely feels the moment the checking stops.

Is relying on AI under pressure a sign of poor judgment?

No. It is a conditions problem, and the research points at conditions, not character. Trained experts with the right answer changed it to match the machine more often when the clock was on. Under uncontrollable stress, flexible thinking narrows and the simple, rehearsed response wins. Reading the first paragraph and sending is the rehearsed response. The judgment is still there. What changed is its cost on that day, and cost can be lowered.

Does asking AI to explain its reasoning reduce over-reliance?

Not reliably. A 2023 study led by Mor Vered found that explanations did not reduce over-reliance and in some conditions increased it, because an answer with a story attached looks more trustworthy. For a leader under load, more text from the tool makes the output look more finished, and finished-looking is exactly what gets waved through. A better safeguard is one specific check chosen before the output arrives, rather than more output to read.

What is the fastest way to check AI output when there is no time?

Choose one check before reading, not after. "Does the hard sentence still say what it said?" "Does the number match the source?" "Is there a line my team would laugh at?" One specific check costs seconds and does not require careful reading of the whole piece. It works because the decision about what matters gets made while the leader still has capacity, not while reading a confident paragraph with eleven minutes left.

Should executives use AI less when they are under pressure?

Using AI less is not the move, and the research does not support it. In the 2025 pathology study, the AI improved overall performance; the losses were specific moments where a right answer got overturned. The goal is fewer of those moments, not fewer tools. Those moments have an address: high load, complex task, running clock. Name the conditions, choose one check, and say what part of the work is the leader's own decision. The tool stays. The leader stays in front of it.

What does AI over-reliance cost a leader over time?

Beyond the visible errors, it costs practice. Checking an output against what you know is a skill that softens when unused. A leader who spends a hard quarter accepting outputs arrives at the next slow week a little less practiced at the thing she used to do without thinking. There is also a team cost: when the leader at the top waves things through, the building learns that waving through is the standard. Leaders are a thermostat, not a thermometer.

Before the next hard Thursday

The next question is Thursday's: what does a team hear when a leader answers under stress, before the words land?

Leadership under pressure is learned in a room, not a paragraph.

Register for The Pressure Room by Sahar Andrade. Thursday, November 19, 11 AM Pacific, free, live on Zoom: https://www.saharandrade.com/pressure-room

About the author

Sahar Andrade, MB.BCh, is a physician turned Neuroleadership Coach, a Forbes Coaches Council member, and the host of AI Café Conversations, a top 2% podcast for executives. She teaches leaders the neuroscience of leadership under pressure through her proprietary B.R.A.I.N.™ framework, five evidence-based principles that create the conditions for lasting leadership change. Connect on LinkedIn: https://www.linkedin.com/in/saharandradespeaker

Everything here comes from my experience and my opinion as a coach. It is not a diagnosis and it is not a fact about you.