Something else: are improvements expanding ZoC to blame for AI unleashing monsters so frequently?
I imagine AI can easily be taught (and probably already has been) not to settle in a spot that will immediately unleash a monster. However, with autoplacement, it seems impossible to prevent your ZoC from accidentally expanding over a lair that was previously safe. I often find myself manually placing improvements away from a dragon just to avoid waking it, which isn't something that the autoplacing AI is capable of.
If ZoC expansion is controlled by specific buildings only, AI can see exactly where the ZoC will expand when queuing up a monument or town hall, and make a decision not to build those if it is afraid of unleashing a monster it can't handle.
If improvements do not expand ZoC, you might find in a bad situation: As the fog of war is just at the next tile, a monster there might attack the city via improvement on the next turn and you will not realise until it is too late (the "there are enemies in your ZoC" message will not appear).
This seems like the most important point to me. One suggestion for mitigating this is to increase the initial city ZoC radius to two (matching that of outposts); this means it will not be a problem for the vast majority of situations. If I still chose to build to the edge of that ZoC, I am ok with being attacked without warning; after all, I am effectively building houses on the edge of civilization, and if I build my house within walking distance of a big dark forest I should not be surprised when I am attacked by a wolf in the middle of the night.
If you want an advanced warning system, build an outpost. Isn't that exactly what outposts are for, historically?