The idea I have (again, perhaps wrongly) is approximately that harmful-to-humans mutations make up a smaller fraction of the space of viable (i.e. allows the organism to survive at all) natural mutations, and a larger fraction of the space of viable transgenic mutations. A secondary point is that this is because the space of viable transgenic mutations is not constrained by the prior state of the genome, the same thing that makes transgenic mutations powerful also makes them risky.
So, I am (or I was, anyways) positing _fewer_ dangerous natural mutations, not necessarily zero.
I'm certainly less convinced that there's an important difference than I was yesterday, though, but mostly because the _accessible_ space of transgenic behaviors is evidently limited by investigator effort, which does not scale, so even if the _theoretical_ space is larger, it doesn't matter, we're not going to sample it.
I don't know, the whole group of flowering plants is about 300 million years old (says Wikipedia). Perhaps you mean 30 million years.
Apparently is a very very old feature (even older than what I though).
In particular:
> The acute oral lethal dose of cyanide for adult humans is reported to be 0.5–3.5 mg/kg (0.2–1.6 mg/lb) of body weight (approximately 50 bitter almonds), whereas for children, consuming 5–10 bitter almonds may be fatal.
The mutation that makes the bitter almonds produce cyanide is very old, totally natural and dangerous. It was mutated at least 300 million yeas ago, before the almond became a different plant than peach and cherry and other similar plants. Probably to make the animals avoid the plant and kernels.