I'm guessing this is your answer:

If the AI knows that you are the aggressor, then it should be expecting an invasion, especially if you are of comparable or greater strength.  It should put out recon parties to detect your approach.  It should anticipate what cities you are likely to target based on its knowledge of which cities you have spotted and determine which cities are more important to defend based on output and proximity to your territory.  In general, the AI should not try to pursue a war on multiple fronts.  If a stronger aggressor declares war on them when they are already at war, they should try all means to end one conflict, or gain allies in their conflict to even the odds.  If they are outmatched, focus on defense and make the aggressors pay for every inch of ground.  Spawning a group or two to hit you from behind in order to distract and delay your attack would also be wise.

As far as attacking goes, the AI, both individually and as allied groups, need to be able to identify targets, assemble forces and time simultaneous attacks.  Here is where it would be nice if the tactical battles could incorporate units in adjacent squares.  Two (or more) allies converging to cooperatively attack your strongest armies or cities would be very evil.  You would almost need another level of strategic AI for alliances above the individual player's decision making processes.  Diversionary tactics are also very useful.  Attack city A to draw forces out of position, then move in the main forces to attack city B.  Of course, the AI should be able to recognize a diversionary attack when it comes and not overreact by moving all its forces in response.

Tactical strategies are also important, as a superior force can still lose if they do not use their forces well.  Minimum, teach the AI to try to not move their units into attack range of the player at the start of a close battle, especially since counter-attacks are not in FE.  Hold the faster moving units back until first contact is made if rushing in would get them torn to pieces before the heavier units can pin the enemy down.  Spread out or mix in with the enemy if AoE spells are in play.  Hold the line if your ranged attacks and spells are superior.  Close in if they are poor.  Either attack to take out enemy damage dealing capability, or attack to inflict losses (kill someone completely ASAP).  Don't just attack whomever is the closest.  Withdraw heavily damaged units if the battle can be won without their help to prevent unnecessary losses.  A well fought battle is the icing on a well planned attack.

As for the diplomatic incentives, perhaps add temporary negative relations for capturing a city.  Say, -3 for a close relation, -2 for a positive relation, -1 for a neutral, +1 for a shared enemy.  And have them stack for each city.  Also, additional, longer term negative relations for razing a city.  They would have to wear off over time, though, or you would have everyone at war with every one fairly soon.  But this would quickly turn the world against a steamroller.  How about negative bonuses for being the highest on the power scale:  -1 for being the highest ("Your power is worrying to us"), -2 for being 20% above the next highest ("Your power is getting out of control"), -3 for being 50% above the next highest ("You must be stopped before you destroy us all!")  There would need to be positive balances, like gifting capitol or resources to produce a mollifying effect: (+3 "You have given us a great gift!")(+2 "You are very generous")(+1 "Your token is appreciated")

Indeed, it is testing like this that caused Civ 5 to become a war-only type game, obsessed with maximizing military production asap rush rush rush.
But i'm sure Stardock tests it in all kinds of ways, this is just 1 little stress test to see how the AI deals with stuff.

Gimmi .78 already!

What if the 'human player' would collect a series of automatic 'opinions' on the AI you want to attack just like the AI does in order to decide what its opinion is of other players?  What if you then had a diplomatic option to 'denounce' the target civilization to other civilizations, and based on their opinion of the target combined with your 'evidence', you could receive a reaction as to how they would feel about you going to war with the target? Maybe this could even reduce their opinion of the target to the point where they are ok with you attacking?  Or even cause them to join in, or break treaties if the evidence is strong enough and lowers their opinion enough?  That way, the system kind of makes use of what is already there in the AI 'opinions'.

Sounds interesting, and would be a good use for diplomatic capital!

(-6.0) Empire razes a Kingdom Capital City - an event that unites the Kingdom sovereigns against the Empire threat.

(-6.0) Kingdom razes a Kingdom Capital City - Any Kingdom sovereign that would do this is evil and must be stopped by the other Kingdoms (not Sovereigns)

(-1.0) Kingdom razes a Kingdom Non-Capital City (not outpost) - causes other Kingdom Sovereigns (not Empire Sovereigns) concern.

(-12.0) Kingdom or Empire razes all cities of another Kingdom striking fear in all the other Kingdom sovereigns (not Empire Sovereigns) that they might be next.

A good diplomacy system doesn't consider the relations between the factions in isolation. Having a system where an action against one faction affects the whole system is better. I was going to say more realistic, but as this is a game about magic I don't know if that is the right word to use here...

I do understand that the idea I presented goes probably too far. It has the risk of making diplomacy too central part of the game. If a really good implementation of the "council" would be done, maybe it would be actually fun to adjust how much you are going to bribe each faction to make them look the other way. But realistically it is unlikely such an implementation is actually possible.

It's fun thinking about these things, though. I don't see a reason to not discuss about these things, as long as we are not claiming "do this or the game is broken". This is about entertainment after all. And there is always the odd change that once in a while something implementable is actually found.

add strategic infiltration with spies or hidden or teleported units operations?

It's a good thing to suggest whatever comes into your head. It's a great thing to put those ideas in a separate post so that dev blogs don't go off topic into the realm of wishful thinking. 

One thing to think about is that we can create our own resources fairly easily. You could even have several types of diplomatic capital that do specific things. And it can all be produced from the current diplomatic buildings. 

The AI, I'd want it to be unpredictable in how it dealt with the early game.  An AI that prepared for a rush and went hybrid would be cripped if you just built.

An AI that rushed would wipe out that turtling.

An AI that built would destroy a half-hearted rush but get wiped out by a true one.

These gambles occur in MP games, the AI can simulate this.  Another option is the AI trying to ally with another player (or you) early one to jajaja another player.

I remember dialog during EWOM about rushing. I think the solution of city militia is a good one and can be further balanced by increasing the quantity and/or strength of the early game militia.

Early rushing by any player should also give any other factions of that race in the game some bonus population, or maybe trigger a "Refugees" event which gives a bonus population growth for a time.