No individual in real life takes actions against 10⁻⁵ events.

(At least no one who is free from mental illness -- OCD for eg.)

That's more like a 10⁻⁶ death risk, since the event risk appears to be below 10⁻⁵ and most don't die of cerebral venous sinus thrombosis.

A one-in-a-million chance? Really? You'd be doing nothing else. Consider flooding of your home, a real risk to life. We take action to avoid that ... at the 10⁻² *annual* risk level, your 1 in 100 flood.

That's above 10⁻¹ *occupation time* risk, if you stay there 10 years. Yet your agent, and your local Council, will tell you that is 'flood free'.

Ok, probably you don't die when the flood actually comes, so the death risk here is maybe only ~10⁻⁴, but there are other serious consequences and few benefits.

Humans are remarkable risk computation engines; it's an evolutionary survival thing. But we are programmed to do that rapidly, via heuristics; we do not do calculations. So when presented with very small numerical risks, we get it all badly wrong.

What is 'safe'? There are things that are impossible, cannot physically happen, where the risk is exactly zero, but they are uninteresting. In the real world 'safe' is a relative concept; it means 'very low risk'.

How low? I think it means 'low enough that a reasonable person would simply ignore the risk'. That I suggest mostly happens at much higher likelihoods than your 10⁻⁵ or 10⁻⁶ level.

But there is an important caveat. What people will accept depends strongly on whether a risk is *imposed* or them by some authority, or instead is knowingly and willingly accepted. Several orders of magnitude difference.

(There is a literature on this stuff; no need to guess.)