AI are not problematic because they're both nonhuman and smart; they're problematic because they're nonhuman, smart, and unpredictable. Lethally so.
Randomly self-modifying AI, yes. Which seems to be a possibility in the setting of EVE. Other than that - for example, as long as the AI can be limited to some operating parameters it cannot circumvent, then they're not that much unpredictable. For example, mathematically speaking the evolutionary computing methods are just elitist multistart random searches. Of course, the randomness is what makes it possible for such methods to find for example alien-looking but effective antenna designs (also the fact that unless programmed to do so, computer "designers" don't have a concept of "beauty" whereas human designer usually do prefer symmetry etc.).
Quantum computers, or computers whose output is modified by random input (for example nuclear decay) can be unpredictable, though, within their programmed limits. (So the Rogue Drone incident would be easily explained by a self-modifying AI being allowed to rewrite a few rules too much, or lacking constraints to its optimization.) Most AI could be expected to be able to consider much more parameters to any given problem compared to humans (who generally manage 7 +/- 2). This means that even IRL AI decisions can be (at a glance) uncomprehensible, though (within the programmed limits) they are perfectly predictable.
Therefore, if for example a Rogue Drone AI chip related to the drones being observed can be captured and possibly used as a processor in feedback loop or the inference engine and ruleset reverse engineered, then their behaviour can be simulated with reasonable accuracy, as long as there's enough computer time (which for example EVE ships have a plenty). The Rogue Drone have nonhuman optimization used on some of their parts (some of it possibly being quite nonoptimal), but for example the most hulls are still following the original designs. (Innards may be swapped, but there's limits to what the drones can do.)
The few existing near-sentient or sentient drones are different, though, specially if they have randomly rewritten their optimization goals.
