Every precaution introduces a risk of its own somewhere else

A countermeasure is an intervention

A precaution is a change to a system, and changes have effects beyond the one intended. Adding a screening step adds the consequences of false positives. Adding a barrier adds whatever happens when the barrier has to be got past in a hurry. Requiring an approval adds delay, and delay has its own casualties in any process where timing matters.

None of this argues against precautions. It argues that the comparison being made is almost always the wrong one: the original hazard against zero, when it should be the original hazard against the hazard of the modified system. The second comparison is harder and it is the only one that answers the question.

Substitution moves risk rather than removing it

The clearest version is when people respond to a restriction by doing something else that carries its own exposure. Discourage one activity and the activity that replaces it is rarely risk-free; make one route difficult and traffic moves to another. The total is what matters, and the total is measured across the whole system rather than within the part that was regulated.

This is not the same as the familiar observation that people behave less carefully when they feel protected. That concerns the same activity being conducted differently. Substitution concerns a different activity being conducted instead, and it can operate even where behaviour within each activity is entirely unchanged.

Worst-case reasoning has two worst cases

Arguments for precaution often work by asking what happens if the hazard is real and nothing is done. That is a legitimate question. What makes it an incomplete argument is that the mirror question is equally legitimate: what happens if the hazard is small and the precaution is imposed anyway.

Both branches have tails. Applied consistently, the principle of acting on the worst case cannot decide anything, because it recommends action and inaction simultaneously. What actually resolves such cases is a judgement about relative magnitudes and about who bears each cost, and it is better to make that judgement explicitly than to let it hide inside a slogan.

Why the introduced risk is systematically underweighted

Several ordinary features of attention push the same way. The original hazard is the thing under discussion, so it is described in detail while the replacement risk is not described at all. The original harms are often identifiable and the new ones statistical. And harms that follow from having acted are counted differently from harms that follow from not acting, which is a well-documented asymmetry in its own right and needs no repeating here.

There is also a measurement asymmetry that outlasts the decision. The prevented harm gets counted, because somebody is looking for it; the induced harm appears in a different category, in a different system, often to different people, and frequently is never attributed to the precaution at all. Absence of an attribution is not absence of an effect.

Distribution is usually the real disagreement

When two people argue about a safety measure while agreeing on all the facts, the disagreement is often about who ends up carrying the residual. A precaution that shifts risk from a visible group to a diffuse one may reduce the total and still be contested, reasonably, by whoever now carries more of it.

Making that explicit tends to shorten the argument. Naming the group that bears the remaining exposure converts an unresolvable dispute about whether something is safe into a discussion about a trade that people can actually take positions on.

It also changes what counts as evidence. Once the question is who carries the residual rather than whether the measure works, the relevant facts are about exposure and circumstance rather than about the mechanism of the hazard, and those are frequently held by the affected group rather than by whoever designed the precaution. That is a practical reason to ask them, quite apart from any argument about fairness.

Comparing honestly without pretending to precision

The minimum standard is that both sides of the comparison are described in the same terms over the same period. If the hazard is stated as an annual chance, so should the induced risk be; if one is stated per exposure and the other per year, the comparison is not a comparison.

The other requirement is to state what is unknown. Frequently the induced risk cannot be quantified at all, and the correct thing to do is to say that rather than to leave it out, since leaving it out is arithmetically identical to setting it to zero. A stated unknown at least remains in view when the decision is revisited.