Pathfinding isn't a problem. I mean its not that hard to create an algorithm that traces through portals. (Big O of N+XN where X is the number of portals in walking range, thus expected to be a rather low number) Its other issues that relate to it in terms of AI trying to judge distance. When it is no longer "can this unit reach this location in 10 steps?" is when its a problem. (that would be just starting on the unit, and doing a recursive loop to check all spaces that are within 10 movement that hasn't already been checked by one of the previous recursive branches)
I mean in theory you could go through each city, and cross reference it with each enemy unit to see how far away each enemy unit is (thus pathfinding through portals to see if the unit is 20 steps away, for example) but then you're talking crap like Algorithms with Big O = N (squared)... which is very CPU cramping (and likely the way I'd do it if I were coding game AI... but thats why I'm not allowed to code AI or collision detection for anything I do anymore).
And lets not forget "checking to see if this particular opponent can even detect the enemy in that space" which could include checking to see if allies or friendly opponents might like the AI player enough to tell them "hey, there's a dude sneaking up on you who's likely not visible to you yet but he's commin' through the portal... you better watch out" and can see the unit for the AI in question.
It gets even rougher when it has to run checks like "does this enemy have access to the planar gate spell, which allows it to jump across maps in certain ways" because then it has to run a check for each of those... Big O = N (cubed), so the enemy turns now take 3-5 min to process enemy movement. Its a freaking chess program, except with 16 opponents rather than just 1.
BUT! then again, this game has a multiplayer component, Gal Civ didn't. The AI doesn't need to compete with human players for understanding portals, other players do. The single player campaign can just be limited on the number of portals, or have portals only used/placed in certain ways that the AI does understand it. Skirmish should just be "whatever is fun" and the player totally customizes that anyway... just make one of the defaults "no portals", and have tons of portals for the "non-cannon" multiplayer settings. Frogboy promised an AI that could handle mods, and pathfinding through portals is certainly needed, even if this game doesn't include it in its cannon, just because that would be used in countless situations that might not even be directly "portals" but just similar in terms of sub-mechanics. AI for portals isn't needed to fulfill that promise, since it isn't something so dependent on the basic functions and compatibility of the engine...