If a mad scientist mastered the technology and introduced a gene drive to make humans dislike broccoli, say, how long would it take for us to detect the attack?

Could we undo it? Since CRISPR modifications have side effects, would we feel safe trying to undo it? Where would we draw the line?

If the mad scientist targeted broccoli bitterness, maybe we would let it pass, especially if we detected the attack after hundreds or thousands of people had been affected. The cure could be more dangerous than the disease.

If the mad scientist decided to try to give everyone blue eyes, maybe we'd act more aggressively to undo it?

I don't think the answers are obvious. And since these are not billion-dollar experiments, it could be a real concern.

As your parent comment suggested, the human generation cycle is not measured in days and it would need hundreds of years to manifest such a trait in the population. Also, this impacts testing as well, so development cycles will be slow too. What you are probably thinking of is a virus infection that proliferates a certain trait. However within 2 or 3 virus generations, evolution will discover that the virus is better off without that trait and will get rid of it. So your mad engineer will not achieve much.

> Could we undo it? Since CRISPR modifications have side effects, would we feel safe trying to undo it? Where would we draw the line?

We do not even know how to do what you are proposing, so why should we know how to undo it? The only answer here I can give you is that we do not seem to see gene drives in nature a lot. If such a powerful tool would exist, evolution (aka selfish genes) would certainly want to use it but at the same time become very inventive to find strategies against it. For example, endogenous retroviruses are a thing, but we still do not see them in action much.