[Twan]

Real AIs (AIs who learn after a mistake and adapt to your strategy like a chess bot) are very rare even in big budget games. So I think you shouldn't expect the dom AI not to repeat the same errors. Games AIs don't think like humans, and the dom AI certainly don't have a memory to remember past failures to take a province or past assassinations.

Anyway to make assassinations less effective, the AI may be coded to split armies between as many commanders as possible each time several commanders moving to the target province are available, or may be coded to evaluate more efficiently province defence ; but don't expect it to ask itself "why did I've failed" then adapt its strategy.