Bingo! Give the man a cigar!I think you get to the point where the Firaxis analyst/programmer knows about the bug/exploit but it's too complex or time-consuming (viz. uneconomical) to make the program changes.

Over the years, we've had some interesting talks with Firaxis, and have learned some things. For instance:There is no publicly available written list of what the AI is programmed to do or not do! So, how do we know what the AI can do?
- The AI Army "bug" is indeed a defect in the code. IIRC, it was introduced after beta testing, or we testers would certainly have caught it.
- The AI knowing the entire map is a "cheat" that was deliberately programmed in (or, at least, not programmed out) to give the AI a bit of "help". Of course, if you know the AI knows this, you can manipulate the AI.
- The turn sequence is much more complex than the average player realizes. It is not a simple "I do everything, then the AI does everything". A search of the forums might turn up some discussion about this. I don't *think* the AI can rush a defender outside the correct sequence, although it may seem like it to the human player.
- Trade costs between the AI are specifically-coded bonuses (or "maluses" (= negative bonuses)) to give the player more of a challenge at higher levels, and an easier time of it on lower levels.
- At no time does the AI check to see if the Civ involved in negotiation is human, except to determine things like trade bonuses. E.g., it checks its internal table of reputations, attitudes, etc. up front for all negotiations, whether player <--> AI, or AI <--> AI, and afterwards only checks if the other side is human in specific situations, like for trade bonuses. So you get the same response as any AI civ would get, if it were in the same circumstance.
Hope this helps.
