Yeah, there was a clever design to measure the harms of smoking while pregnant.

Obviously, you can't just assign one group of women to a treatment group and have them smoke two packs a day.

What you can use in these situations is an "intent to treat" strategy. In that study, one group of women received some extra literature and counseling about the harms of smoking. That group of women, who previously reported the same smoking rates as the control group, smoked less during their pregnancies. By attributing the reduction in smoking to the intervention (and assuming there was no other channel by which that intervention impacted health) you can treat the difference smoking as though it's exogenous.

In other words, it's almost empirically equivalent to actually assigning one group to smoke more, without being morally equivalent.