The man who never collapsed.
In 1991, psychologist Paul Salkovskis described a puzzle that had been bothering anxiety researchers. A man repeatedly felt his legs go weak and became convinced he was about to collapse. Each time, he gripped whatever was nearby, tensed his legs, and found a seat. He never collapsed — and yet years of never collapsing hadn't reduced the fear at all. Salkovskis' insight was that the gripping was the problem. Because the man always did it, every non-collapse got credited to the gripping rather than to the truth: the sensation was harmless. Salkovskis called these "safety-seeking behaviours" — things done to prevent a feared catastrophe that, by always being there, quietly steal the credit for the catastrophe never happening. The fear survives every single near miss it manufactures.
Drop the armour, lose the fear faster.
This wasn't left as theory. In 1995, Adrian Wells, David Clark, Salkovskis and colleagues took people with social anxiety through two exposures to a feared social situation: one done as usual, one with their safety behaviours deliberately dropped — no rehearsing sentences, no gripping the glass, no avoiding eye contact. One undefended session reduced anxiety and belief in the feared outcome more than ordinary exposure did. A 1999 study found the same in panic: fifteen minutes of facing the fear without the props beat fifteen minutes of facing it with them. The pattern explained a long-standing mystery — why people can be "exposed" to social life daily for decades without improving. They were present, but never undefended, so nothing was ever truly tested.
Protection that manufactures danger.
The reverse experiment is the unsettling one. In 2008, Brett Deacon and Danielle Maack asked people with no particular contamination fears to perform contamination safety behaviours for a week — disinfecting hands after door handles, avoiding "dirty" surfaces. By the end of the week, their threat estimates, contamination fear and avoidance had all measurably increased. A larger, better-controlled experiment by Eva van Dis and colleagues in 2022 found the same in miniature: simply performing safety behaviours toward a harmless object pushed up people's sense of threat. Safety behaviours, it seems, don't just preserve existing fear; they can feed it, because the behaviour itself reads as a danger signal. If I'm doing all this protecting, some part of the mind concludes, there must be something to protect against. The armour doesn't just hide the evidence of safety. It quietly testifies that the world is dangerous.
Reassurance is the portable version.
Asking "are you sure it's nothing?" works exactly the same way, which is why Salkovskis and Hilary Warwick identified reassurance-seeking as a compulsion in all but name back in 1986. The defining feature, per later work by Adam Radomsky's group, is that you're soliciting safety information you already have. The answer brings relief; the relief reinforces the asking; the doubt returns sooner and stronger; and each round teaches your threat system that the question genuinely needed answering. Crucially, the same act can be healthy or unhealthy depending on its job — researchers Richard Thwaites and Mark Freeston showed you can't classify a behaviour by watching it, only by asking what it's for. Slow breathing because it's pleasant is coping. Slow breathing to prevent suffocation is a safety behaviour — same lungs, opposite lesson.
Sleep has its own collection.
Insomnia runs on safety behaviours, as Allison Harvey's cognitive model of insomnia laid out: going to bed early "to catch up", lying in, napping, cancelling things after a bad night — each one diluting sleep pressure and deepening the conviction that sleep is fragile. The best-proven example is clock-watching. Nicole Tang, Schmidt and Harvey randomised people to monitor a clock while falling asleep or not: clock-monitors worried more, took longer to fall asleep — even the good sleepers — and overestimated how bad their night had been. The 1am arithmetic ("if I fall asleep right now I get five hours") is reassurance-seeking aimed at a clock, and it backfires identically. Even a wind-down routine can cross the line: used because it's pleasant, it's a tool; clung to as the only thing standing between you and a catastrophic night, it has joined the armoury.
Taking the armour off — gradually counts.
The way out is not white-knuckled bravery but curiosity. First, map the armour: list what you do before, during and after the feared situation, and ask of each item, "what do I believe would happen without it?" If the answer is a catastrophe, that behaviour is holding the fear in place. Then test, don't endure: write the specific prediction, drop one prop, and check what actually happens — that's the engine of the 1995 and 1999 studies. Fair warning from the research: dropping needn't be absolute on day one. The experimental record is genuinely mixed on whether keeping a safety behaviour always sabotages exposure — a 2016 meta-analysis by Ann Meulders and colleagues found the evidence inconclusive — and work by Jack Rachman and Adam Radomsky found that a temporary crutch which gets you in the door can be fine, provided it is deliberately faded rather than becoming permanent equipment. The settled part is the clinical consensus: identify the props, and over time, let them go. For reassurance loops, delay the question twenty minutes and allow one answer per genuinely new question. And for entrenched anxiety, health anxiety or OCD, this is core CBT territory where a therapist makes the work faster and safer.