You could even make the question more precise. If you market a product where you know it could be safe, but also know X% of the users will accidentally kill themselves (and maybe others), then for what X is it unethical? I think there are probably values on each end where it's clearly either. 100-epsilon is bad. epsilon is fine. Somewhere in between it gets fuzzy.

Another interesting question is that when accidents come up, do you get to blame those X% of users for failing, even though at the design stage you already knew and decided they would fail and be killed.

It's really more of a trolley problem, in that it's interesting to think about, but in reality there's more context and circumstances that make specific cases clearer.