How the body's oldest defence dismantles our best teams, and what safety really buys us.
In brief.
Defensive and reactive behaviour at work is a survival response, run by threat circuitry that cannot tell a hostile boss from a hostile predator. Under threat the thinking brain goes partly offline, and over time the wear shows up as burnout and ill health. Psychological safety is the condition that switches the alarm off, which is why it consistently tracks with wellbeing, learning and performance. This piece connects the neuroscience of threat to the organisational science of safety, and argues they are the same story told from opposite ends.
Early in the 1990s, a doctoral student named Amy Edmondson went looking for an unremarkable result. Studying medical teams across two hospitals under the Harvard psychologist J. Richard Hackman, she expected the obvious. Better teams, with better leadership and relationships, would make fewer mistakes. She had survey data on team quality and independent records of medication errors. All that remained was to line them up.
The correlation came back the wrong way round. The better teams appeared to be making more mistakes, not fewer.
Then she had the thought that would shape the next thirty years of her work. The good teams were not making more errors. They were willing to talk about them (Edmondson, 2023). Where a team worked well, a nurse could say "I think I've just given the wrong dose" out loud and expect problem-solving rather than blame. In the weaker teams the same nurse kept quiet. The errors happened in both places. Only one kind of team could bear to look at them.
Edmondson named what the good teams had built: psychological safety, "a shared belief held by members of a team that the team is safe for interpersonal risk-taking" (Edmondson, 1999). When Google later set out to find what made its own teams effective, an effort called Project Aristotle that combed through 180 teams, the strongest single answer was the same one (Duhigg, 2016). To see why it matters so much, it helps to go down into the oldest part of the brain, to the flinch.
The two-hundred-millisecond veto.
Deep in each temporal lobe sits an almond-shaped cluster of neurons, the amygdala, whose job is to answer one question faster than anything else the brain does. Is this dangerous? The neuroscientist Joseph LeDoux showed that a threatening signal can reach it by a fast subcortical route and trigger a defensive response before the thinking regions have decided what is happening (LeDoux, 1996; 2000). The body braces before the mind rules.
Daniel Goleman popularised this as the "amygdala hijack" (Goleman, 1995), though LeDoux presses for a correction. What he studies is not "fear" but the brain's defensive survival circuits, the machinery that detects a threat and answers it; the conscious feeling comes later and separately (LeDoux, 2015). The circuit does not wait for a predator. It fires for a raised eyebrow, a curt reply, not enough likes on LinkedIn, the silence after you admit you do not understand something. Our threats are social now, but the hardware cannot tell a threat to the ego from a threat to the throat, and runs the same programme for both.
Taking the pilot offline.
The prefrontal cortex, behind the forehead, is where the good work happens: holding ideas in mind, weighing them, governing impulse, saying the difficult true thing. It is also fragile under threat. The Yale neuroscientist Amy Arnsten has shown that even mild, uncontrollable stress floods it with noradrenaline and dopamine that disconnect its networks, taking it offline and handing control to faster, habitual circuits (Arnsten, 2009). Sustained stress makes the harm structural, with prefrontal branches retracting while the amygdala's grow (Arnsten et al., 2019).
The conditions that make a workplace feel threatening therefore switch off the hardware we most need people to be running. Threat does not only make us unhappy; it leaves us less able to think, learn and invent, narrowing the range of thought to a pinpoint (Fredrickson, 2001). The nurse who kept quiet was not a coward. That was physiology doing what it evolved to do.
The wear on the machine.
One flinch is cheap. The body braces, the threat passes, the systems reset. The trouble starts when the threat never passes. Bruce McEwen named the cost allostatic load, the wear from stress systems that switch on too often or never fully off (McEwen & Stellar, 1993; McEwen, 1998). Cortisol that protects in a crisis turns corrosive when the crisis has no end, and sustained load is tied to cardiovascular disease, weakened immunity, impaired memory, depression and faster ageing (Juster, McEwen & Lupien, 2010). A chronically unsafe workplace is a physiological tax, paid in the end in health.
This is where the organisational numbers stop looking like coincidence. In a 2026 study of 821 health and social care workers, team psychological safety correlated with wellbeing at r = .68 and, inversely, with burnout at r = −.59 (Cogan, Smith & Deakin, 2026). A meta-analysis across 136 samples and more than twenty-two thousand people links it to engagement, learning and creativity, and away from withdrawal (Frazier et al., 2017). Most of this evidence is correlational (CIPD, 2024), so causation is not proven from the org data alone. What lifts it above coincidence is the neuroscience: we already know how a threatening climate grinds a person down.
The clever people learn worst.
Long before brain scans, the Harvard theorist Chris Argyris watched what defensiveness does to how an organisation thinks. Under threat, he found, almost everyone falls back on a self-protective default: win, stay in control, avoid embarrassment (Argyris & Schön, 1974). The result is what he called defensive routines, "actions or policies that prevent individuals or segments of the organisation from experiencing embarrassment or threat, while simultaneously preventing them from identifying and reducing the causes" (Argyris, 1990). They seal themselves shut, because the thing that cannot be discussed includes the fact that it cannot be discussed.
This is where an organisation stops learning. Argyris separated single-loop learning, correcting an error while leaving your assumptions intact, from double-loop learning, which questions the assumptions themselves (Argyris, 1991). Defensiveness allows the first and forbids the second, because rethinking means being wrong out loud, the very event the flinch exists to prevent. His sharpest observation was about who does this worst: not the strugglers but the stars, clever people who have rarely failed and so never learned to digest it, and who become the most defensive in the room the moment they cannot master something at once (Argyris, 1991). The defensive routine is the amygdala's flinch scaled up to a whole company.
What safety actually buys.
We tend to talk about psychological safety as niceness, as lowering the bar. Edmondson is firm that it is the opposite: it does not remove standards, it makes high standards survivable, and it is built to sit alongside real accountability (Edmondson, 2019). What it does, read through the flinch, is neurological. It lowers the ambient threat enough that the defensive circuitry stops firing at shadows, which keeps the prefrontal cortex online, which lets the nurse say the sentence and the second loop close. Safe teams do feel better, but the deeper point is that they can keep thinking in conditions that switch everyone else's thinking off.
The instinct that undoes our teams is not a defect to train away. It is the oldest and most loyal part of us, still guarding against predators that have largely gone, unable to tell the meeting room from the savannah. That is not its failure. It becomes ours if we build organisations that keep it needlessly afraid, and our opportunity if we build the one thing that lets a frightened animal lower its guard and begin, at last, to learn.
None of this is a training problem. The flinch does not respond to being told about itself, and a leader who knows their pattern still runs it the moment the pressure lands, because the circuitry was built to move faster than knowing. What changes it is reaching the structure underneath, the thing the pattern is protecting, so it loses the reason it was firing.
That is the work we do, in ninety minutes, with the leader whose behaviour sets the temperature for everyone around them. Their team feels the difference within weeks. The cost of all that guarding has been sitting on your P&L the whole time.