The principle says that action should not wait for proof of harm where the harm would be severe.
Stated that way it is almost and almost empty.
Every action has risks, and so does every refusal to act, which the principle in its strong form ignores.
to license a medicine avoids the risk of side effects and accepts the risk of untreated disease.
A principle that counts only one of those is not cautious but selective.
The selection has a predictable direction: harms caused by action are weighted more heavily than harms caused by inaction.
This is a well-documented feature of human judgement and is not obviously a virtue.
A regulator who approves a harmful drug is blamed; one who delays a drug is not.
The patients who died waiting are as dead as those who died from a side effect and are statistically invisible.
A weaker version of the principle avoids this and is the version most usefully defended.
It holds that where a harm would be , uncertainty counts against the option.
Irreversibility does real work here, because it distinguishes the cases rather than covering all of them.
a species and delaying a drug are both harms, and only one of them can be undone.
On this version the principle is a rule about the value of keeping options open.
That is a substantive claim and it can be argued with, which the strong version cannot.
A principle nobody can argue with is generally one that decides nothing.
It will then be by both parties to every dispute, as this one reliably is.
The test of a principle is whether anybody has ever changed their position because of it.